diff --git a/.changeset/bright-docs-integrations.md b/.changeset/bright-docs-integrations.md
deleted file mode 100644
index 49a927239..000000000
--- a/.changeset/bright-docs-integrations.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/integrations": patch
----
-
-Document the public API.
diff --git a/.changeset/bright-lion-nodes.md b/.changeset/bright-lion-nodes.md
deleted file mode 100644
index 45372eafb..000000000
--- a/.changeset/bright-lion-nodes.md
+++ /dev/null
@@ -1,26 +0,0 @@
----
-"@agentskit/adapters": patch
-"@agentskit/angular": patch
-"@agentskit/core": patch
-"@agentskit/cross-platform": patch
-"@agentskit/eval": patch
-"@agentskit/ink": patch
-"@agentskit/integrations": patch
-"@agentskit/mcp": patch
-"@agentskit/memory": patch
-"@agentskit/net": patch
-"@agentskit/observability": patch
-"@agentskit/rag": patch
-"@agentskit/react": patch
-"@agentskit/react-native": patch
-"@agentskit/runtime": patch
-"@agentskit/sandbox": patch
-"@agentskit/solid": patch
-"@agentskit/statechart": patch
-"@agentskit/svelte": patch
-"@agentskit/templates": patch
-"@agentskit/tools": patch
-"@agentskit/vue": patch
----
-
-Declare the CI-proven Node.js floor (>=20.19, the "node (minimum)" job) in package metadata where it was missing, keeping existing floors, and remove confirmed unused development dependencies.
diff --git a/.changeset/bright-sheep-smile.md b/.changeset/bright-sheep-smile.md
deleted file mode 100644
index 6792ba5ce..000000000
--- a/.changeset/bright-sheep-smile.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/eval': patch
----
-
-Reuse the shared core record guard when validating replay cassettes.
diff --git a/.changeset/calm-docs-rag.md b/.changeset/calm-docs-rag.md
deleted file mode 100644
index 5f7ab95f9..000000000
--- a/.changeset/calm-docs-rag.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/rag": patch
----
-
-Document the public API.
diff --git a/.changeset/chat-memory-postgres.md b/.changeset/chat-memory-postgres.md
deleted file mode 100644
index 1422b116e..000000000
--- a/.changeset/chat-memory-postgres.md
+++ /dev/null
@@ -1,7 +0,0 @@
----
-'@agentskit/memory': minor
----
-
-Add the optional `@agentskit/memory/postgres` subpath with BYO Drizzle/node-postgres ChatMemory, exported schema and migration SQL, tenant/session isolation, full message-part persistence, and opt-in turn-aware retention. Connection lifetime remains host-owned for Node pools and Workers clients.
-
-Bounded retention cuts only at user turn boundaries and keeps the leading system prompt, so turns and tool call/result groups are never split and history is never emptied. Supports Drizzle 0.44 and 0.45 (from 0.45.3).
diff --git a/.changeset/clear-waves-fall.md b/.changeset/clear-waves-fall.md
deleted file mode 100644
index 2d5397614..000000000
--- a/.changeset/clear-waves-fall.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/rag": minor
----
-
-Add createPartitionedRAG with construction-bound metadata isolation, namespaced chunk IDs, bounded candidate retrieval, and explicit recall diagnostics. Existing createRAG and the core Retriever contract are unchanged.
diff --git a/.changeset/codex-w02-adapters.md b/.changeset/codex-w02-adapters.md
deleted file mode 100644
index ac77388ba..000000000
--- a/.changeset/codex-w02-adapters.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/adapters': patch
----
-
-Reuse core secret pattern redaction in CLI diagnostics while preserving configured literal-value redaction.
diff --git a/.changeset/codex-w02-integrations.md b/.changeset/codex-w02-integrations.md
deleted file mode 100644
index 60d42156d..000000000
--- a/.changeset/codex-w02-integrations.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/integrations': patch
----
-
-Reuse core secret redaction for HTTP response diagnostics after verifying existing cases remain covered.
diff --git a/.changeset/controller-review-recovery.md b/.changeset/controller-review-recovery.md
deleted file mode 100644
index 41eb11301..000000000
--- a/.changeset/controller-review-recovery.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/core": patch
----
-
-Keep older pending or superseded tool calls from blocking later controller turns, recover decisions after a pending-registration save failure, and resume interrupted streaming messages loaded from memory. Use a fresh run ID for durable decision executions outside a live stream.
diff --git a/.changeset/doc03-adapters-main.md b/.changeset/doc03-adapters-main.md
deleted file mode 100644
index c14ed2f5b..000000000
--- a/.changeset/doc03-adapters-main.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/adapters": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-adapters-subpaths.md b/.changeset/doc03-adapters-subpaths.md
deleted file mode 100644
index c14ed2f5b..000000000
--- a/.changeset/doc03-adapters-subpaths.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/adapters": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-angular.md b/.changeset/doc03-angular.md
deleted file mode 100644
index 9aa64ec40..000000000
--- a/.changeset/doc03-angular.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/angular': patch
----
-
-Document public API.
diff --git a/.changeset/doc03-cli.md b/.changeset/doc03-cli.md
deleted file mode 100644
index 50f1ea1a4..000000000
--- a/.changeset/doc03-cli.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/cli": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-core-root.md b/.changeset/doc03-core-root.md
deleted file mode 100644
index 4af08b5f0..000000000
--- a/.changeset/doc03-core-root.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/core': patch
----
-
-Document the public API of the root entry and shared chat types.
diff --git a/.changeset/doc03-core-subpaths.md b/.changeset/doc03-core-subpaths.md
deleted file mode 100644
index 73c6f4062..000000000
--- a/.changeset/doc03-core-subpaths.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/core': patch
----
-
-Document the public API of the named core export subpaths.
diff --git a/.changeset/doc03-cross-platform.md b/.changeset/doc03-cross-platform.md
deleted file mode 100644
index 9632144b8..000000000
--- a/.changeset/doc03-cross-platform.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/cross-platform': patch
----
-
-Document the public API.
diff --git a/.changeset/doc03-eval-braintrust.md b/.changeset/doc03-eval-braintrust.md
deleted file mode 100644
index e0681ee9d..000000000
--- a/.changeset/doc03-eval-braintrust.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/eval-braintrust": patch
----
-
-Document the public Braintrust scoring, runner, and regression APIs.
diff --git a/.changeset/doc03-eval.md b/.changeset/doc03-eval.md
deleted file mode 100644
index 25cb04aa1..000000000
--- a/.changeset/doc03-eval.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/eval": patch
----
-
-Document the public evaluation, replay, snapshot, diff, and CI reporting APIs.
diff --git a/.changeset/doc03-ink.md b/.changeset/doc03-ink.md
deleted file mode 100644
index 0ea742a4d..000000000
--- a/.changeset/doc03-ink.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/ink": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-mcp.md b/.changeset/doc03-mcp.md
deleted file mode 100644
index 915a68838..000000000
--- a/.changeset/doc03-mcp.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/mcp': patch
----
-
-Document the public API.
diff --git a/.changeset/doc03-net.md b/.changeset/doc03-net.md
deleted file mode 100644
index 6b19e8255..000000000
--- a/.changeset/doc03-net.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/net': patch
----
-
-Document the public API.
diff --git a/.changeset/doc03-react-native.md b/.changeset/doc03-react-native.md
deleted file mode 100644
index a3c259f9a..000000000
--- a/.changeset/doc03-react-native.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/react-native": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-react.md b/.changeset/doc03-react.md
deleted file mode 100644
index 69a54cdb5..000000000
--- a/.changeset/doc03-react.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/react': patch
----
-
-Document the public React API.
diff --git a/.changeset/doc03-runtime-shared-context.md b/.changeset/doc03-runtime-shared-context.md
deleted file mode 100644
index 5d62f0ced..000000000
--- a/.changeset/doc03-runtime-shared-context.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/runtime': patch
----
-
-Document the shared-context API.
diff --git a/.changeset/doc03-runtime.md b/.changeset/doc03-runtime.md
deleted file mode 100644
index f65688d70..000000000
--- a/.changeset/doc03-runtime.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/runtime': patch
----
-
-Document the public API with JSDoc.
diff --git a/.changeset/doc03-sandbox.md b/.changeset/doc03-sandbox.md
deleted file mode 100644
index 47fdbd4ac..000000000
--- a/.changeset/doc03-sandbox.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/sandbox': patch
----
-
-Document the public API with JSDoc.
diff --git a/.changeset/doc03-skills.md b/.changeset/doc03-skills.md
deleted file mode 100644
index 1445ec9dc..000000000
--- a/.changeset/doc03-skills.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/skills': patch
----
-
-Document the public API.
diff --git a/.changeset/doc03-solid.md b/.changeset/doc03-solid.md
deleted file mode 100644
index 59e6fa289..000000000
--- a/.changeset/doc03-solid.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/solid': patch
----
-
-Document public API.
diff --git a/.changeset/doc03-statechart.md b/.changeset/doc03-statechart.md
deleted file mode 100644
index 4e57848a3..000000000
--- a/.changeset/doc03-statechart.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/statechart": patch
----
-
-Document public API.
diff --git a/.changeset/doc03-svelte.md b/.changeset/doc03-svelte.md
deleted file mode 100644
index f6407bea7..000000000
--- a/.changeset/doc03-svelte.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/svelte': patch
----
-
-Document the public Svelte API.
diff --git a/.changeset/doc03-templates.md b/.changeset/doc03-templates.md
deleted file mode 100644
index e903c5df7..000000000
--- a/.changeset/doc03-templates.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/templates': patch
----
-
-Document public API.
diff --git a/.changeset/doc03-tools.md b/.changeset/doc03-tools.md
deleted file mode 100644
index 044e4745d..000000000
--- a/.changeset/doc03-tools.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/tools': patch
----
-
-Document the public API.
diff --git a/.changeset/doc03-vue.md b/.changeset/doc03-vue.md
deleted file mode 100644
index d8f6d1297..000000000
--- a/.changeset/doc03-vue.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/vue': patch
----
-
-Document public API.
diff --git a/.changeset/durable-controller-decisions.md b/.changeset/durable-controller-decisions.md
deleted file mode 100644
index bf76bf38f..000000000
--- a/.changeset/durable-controller-decisions.md
+++ /dev/null
@@ -1,10 +0,0 @@
----
-"@agentskit/core": minor
-"@agentskit/svelte": minor
----
-
-Add durable `controller.decide` with an injectable atomic decision store, persisted confirmation snapshots, terminal replay, and tool-result continuation across controllers. Preserve and deprecate generation-local `approve`/`deny`. Accept `ContentPart[]` in controller `send` alongside legacy strings.
-
-`ChatController.decide` is now a required interface member, and `ChatController.send` and shared `ChatReturn.send` accept `string | ContentPart[]`. Custom `ChatController` implementations must add `decide`; implementations of either interface must widen `send`. Existing string callers continue to work. The Svelte chat store exposes `decide` with the same lifecycle guards as its other asynchronous actions.
-
-Durable decisions preserve unsaved live turns, validate the pending conversation before claiming, and reconcile sibling outcomes without mutating subscriber snapshots.
diff --git a/.changeset/durable-controller-races.md b/.changeset/durable-controller-races.md
deleted file mode 100644
index 92d25fd6a..000000000
--- a/.changeset/durable-controller-races.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/core": patch
----
-
-Preserve live stream chunks, later turns and new confirmations while durable decision storage is pending. Recheck the target before execution and settle a superseded claim without running its tool. Store-backed legacy approvals and denials share the durable claim boundary.
diff --git a/.changeset/durable-cost-store.md b/.changeset/durable-cost-store.md
deleted file mode 100644
index a2169eb28..000000000
--- a/.changeset/durable-cost-store.md
+++ /dev/null
@@ -1,9 +0,0 @@
----
-'@agentskit/observability': minor
----
-
-Add the `CostStore` port (`reserve`, `commit`, `release`, `window`) for durable, tenant-scoped spend accounting, with `createInMemoryCostStore` for development and an optional `@agentskit/observability/postgres` subpath: BYO Drizzle/node-postgres store with atomic cap admission, idempotent reservations, a usage ledger, exported schema and migration SQL. `@agentskit/observability/cost-store-contract` exports the contract suite for custom stores.
-
-`costGuard`, `multiTenantCostGuard` and `createAdvancedCostGuard` accept an optional `store`. With it, each priced call is written to the store and the budget is checked against spend recorded by every instance in the current accounting window (UTC month by default). Without it behavior is unchanged, except for a one-time warning in production runtimes (`NODE_ENV=production`, or Cloudflare Workers with `NODE_ENV` unset; `warnWithoutStore` overrides it), because process-local totals do not hold across instances or isolates. A store write that fails goes to `onError`, or to one warning when no handler is set. `costGuard` also accepts `tenant` to name the store partition.
-
-`expirePostgresCostReservations({ db, olderThan })` releases reservations that were never settled; the store has no background reaper. `reserve` with the id of a released reservation throws `AK_COST_RESERVATION_RELEASED`. Match `CostReservationError` by `code`: each CommonJS subpath bundles its own copy of the class.
diff --git a/.changeset/eval-result-provenance.md b/.changeset/eval-result-provenance.md
deleted file mode 100644
index b145891e8..000000000
--- a/.changeset/eval-result-provenance.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/eval": minor
----
-
-Add the portable `@agentskit/eval/provenance` subpath with canonical SHA-256 result records with suite, case-set, subject and runner provenance, tamper verification, and precision/recall helpers.
diff --git a/.changeset/fair-tools-resume.md b/.changeset/fair-tools-resume.md
deleted file mode 100644
index 96919f30a..000000000
--- a/.changeset/fair-tools-resume.md
+++ /dev/null
@@ -1,7 +0,0 @@
----
-"@agentskit/core": patch
----
-
-Register streamed confirmations on arrival, treat identical in-flight decisions as no-ops, and preserve active streams when a decision is rejected. Resume only the latest completed assistant turn after every confirmation is decided, with results immediately following their originating assistant in call order.
-
-Restarting streamed decisions requires current ChatMemory because immutable arrival snapshots may omit subsequent calls. Without current conversation memory, an incomplete streamed snapshot raises AK_CONFIG_INVALID before claiming or executing.
diff --git a/.changeset/fuzzy-cats-listen.md b/.changeset/fuzzy-cats-listen.md
deleted file mode 100644
index f4c306962..000000000
--- a/.changeset/fuzzy-cats-listen.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/tools': patch
----
-
-Reuse the shared core record guard in the MCP client and server.
diff --git a/.changeset/huge-monkeys-melt.md b/.changeset/huge-monkeys-melt.md
deleted file mode 100644
index 75e73c769..000000000
--- a/.changeset/huge-monkeys-melt.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/integrations": patch
----
-
-Bound provider response bodies in integrations actions
diff --git a/.changeset/lucky-kids-think.md b/.changeset/lucky-kids-think.md
deleted file mode 100644
index 2716b7342..000000000
--- a/.changeset/lucky-kids-think.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/mcp': patch
----
-
-Reuse the shared core record guard when reading registry projections.
diff --git a/.changeset/mcp-metadata-confirmation.md b/.changeset/mcp-metadata-confirmation.md
deleted file mode 100644
index 9e78720e4..000000000
--- a/.changeset/mcp-metadata-confirmation.md
+++ /dev/null
@@ -1,6 +0,0 @@
----
-"@agentskit/mcp": minor
-"@agentskit/tools": patch
----
-
-Expose the host confirmation hook through the MCP server wrapper, retaining default denial for confirmation-required tools. Preserve optional MCP tool titles and advisory annotations across server listing and client import without changing core contracts or protocol negotiation.
diff --git a/.changeset/migrate-to-renovate.md b/.changeset/migrate-to-renovate.md
deleted file mode 100644
index 1c2acbbac..000000000
--- a/.changeset/migrate-to-renovate.md
+++ /dev/null
@@ -1,4 +0,0 @@
----
----
-
-Migrate dependency updates from Dependabot to the organization Renovate preset.
diff --git a/.changeset/net09b-cli.md b/.changeset/net09b-cli.md
deleted file mode 100644
index 702dda9a9..000000000
--- a/.changeset/net09b-cli.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/cli": patch
----
-
-Bound CLI RAG embedding and registry response bodies.
diff --git a/.changeset/net09b-core-sso.md b/.changeset/net09b-core-sso.md
deleted file mode 100644
index 1975ec791..000000000
--- a/.changeset/net09b-core-sso.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/core': patch
----
-
-Bound streamed OIDC JWKS responses before parsing them.
diff --git a/.changeset/net09b-runtime.md b/.changeset/net09b-runtime.md
deleted file mode 100644
index 6987d6909..000000000
--- a/.changeset/net09b-runtime.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/runtime': patch
----
-
-Use abortable network sleep for durable step retry delays.
diff --git a/.changeset/net09b-templates.md b/.changeset/net09b-templates.md
deleted file mode 100644
index 0e102859d..000000000
--- a/.changeset/net09b-templates.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/templates": patch
----
-
-Document the dependency boundary for generated HTTP blueprint source.
diff --git a/.changeset/net09b-tools.md b/.changeset/net09b-tools.md
deleted file mode 100644
index 0130c68f7..000000000
--- a/.changeset/net09b-tools.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/tools': patch
----
-
-Bound document downloads and Slack error response bodies with `@agentskit/net`.
diff --git a/.changeset/public-api-build-scope.md b/.changeset/public-api-build-scope.md
deleted file mode 100644
index 33d76fdf4..000000000
--- a/.changeset/public-api-build-scope.md
+++ /dev/null
@@ -1,4 +0,0 @@
----
----
-
-Scope the public API snapshot CI gate to the affected build plan while retaining full release and baseline update checks.
diff --git a/.changeset/q02-tsup-shared.md b/.changeset/q02-tsup-shared.md
deleted file mode 100644
index 081809b55..000000000
--- a/.changeset/q02-tsup-shared.md
+++ /dev/null
@@ -1,4 +0,0 @@
----
----
-
-Centralize repeated tsup defaults in `tsup.shared.ts`; build output is unchanged, so no package version bump.
diff --git a/.changeset/quiet-adapters-images.md b/.changeset/quiet-adapters-images.md
deleted file mode 100644
index e85c6c2dc..000000000
--- a/.changeset/quiet-adapters-images.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/adapters": minor
----
-
-Serialize supported image/file parts in OpenAI-compatible, Anthropic, Gemini/Vertex and Ollama requests. Preserve text-only model behavior and older multimodal histories by falling back to text parts and short attachment markers, omitting binary data and data URLs, with at most one warning per request when multimodal capability is disabled; OpenAI and Anthropic assistant binary parts also use the text fallback. Unsupported modalities and sources still raise CAPABILITY_UNSUPPORTED when multimodal input is enabled. Explicit capabilities override the expanded model-name heuristic. Add configurable path, headers and fetch, optional Bearer authentication for compatible endpoints, OpenAI-compatible and Cloudflare REST presets, an OpenRouter alias, and explicit catalog URL variables. Existing text-only and default endpoint behavior is preserved.
diff --git a/.changeset/quiet-docs-memory.md b/.changeset/quiet-docs-memory.md
deleted file mode 100644
index 53286511d..000000000
--- a/.changeset/quiet-docs-memory.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/memory": patch
----
-
-Document the public API.
diff --git a/.changeset/quiet-observability-docs.md b/.changeset/quiet-observability-docs.md
deleted file mode 100644
index ddac555f8..000000000
--- a/.changeset/quiet-observability-docs.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/observability': patch
----
-
-Document the public API.
diff --git a/.changeset/sec-advisories-1003.md b/.changeset/sec-advisories-1003.md
deleted file mode 100644
index a845151cc..000000000
--- a/.changeset/sec-advisories-1003.md
+++ /dev/null
@@ -1,2 +0,0 @@
----
----
diff --git a/.changeset/small-clocks-jam.md b/.changeset/small-clocks-jam.md
deleted file mode 100644
index 44f31c7e6..000000000
--- a/.changeset/small-clocks-jam.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@agentskit/cross-platform": minor
----
-
-Add framework-agnostic shared testing helpers.
diff --git a/.changeset/svelte-controller-send.md b/.changeset/svelte-controller-send.md
deleted file mode 100644
index 814e65481..000000000
--- a/.changeset/svelte-controller-send.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/svelte': patch
----
-
-Forward content parts accepted by the core controller through the chat store while preserving destroyed-store rejection.
diff --git a/.changeset/tidy-users-call.md b/.changeset/tidy-users-call.md
deleted file mode 100644
index 8fbf484fc..000000000
--- a/.changeset/tidy-users-call.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/adapters': patch
----
-
-Reuse the shared core record guard in CLI response and manifest parsing.
diff --git a/.changeset/w03-weak-id-rule.md b/.changeset/w03-weak-id-rule.md
deleted file mode 100644
index a81bae357..000000000
--- a/.changeset/w03-weak-id-rule.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/core': patch
----
-
-Extend the `no-math-random-id` rule to catch base-36 random strings and template-built identifiers, tokens, keys and temporary paths.
diff --git a/.changeset/warm-pandas-jam.md b/.changeset/warm-pandas-jam.md
deleted file mode 100644
index 985a915b0..000000000
--- a/.changeset/warm-pandas-jam.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-'@agentskit/eval-braintrust': patch
----
-
-Reuse the shared core record guard when parsing agent results.
diff --git a/.doc-bridge/capabilities.json b/.doc-bridge/capabilities.json
index f0a907e54..44861198d 100644
--- a/.doc-bridge/capabilities.json
+++ b/.doc-bridge/capabilities.json
@@ -7,7 +7,7 @@
     "name": "AgentsKit",
     "root": "."
   },
-  "contentHash": "67a7cb465aadd34c99468f94869e00495bcc92863f5819adaec1eb1e13e13d2c",
+  "contentHash": "7d1e14de1029deaed9354aff320bedf1d83f127f67e70b1f61c406fe5724dc95",
   "artifacts": {
     "index": ".doc-bridge/index.json",
     "llmsTxt": ".doc-bridge/llms.txt"
diff --git a/.doc-bridge/index.json b/.doc-bridge/index.json
index a26e7a7c7..7f7b93c54 100644
--- a/.doc-bridge/index.json
+++ b/.doc-bridge/index.json
@@ -1,8 +1,8 @@
 {
   "schemaVersion": 1,
-  "contentHash": "67a7cb465aadd34c99468f94869e00495bcc92863f5819adaec1eb1e13e13d2c",
+  "contentHash": "7d1e14de1029deaed9354aff320bedf1d83f127f67e70b1f61c406fe5724dc95",
   "contentHashAlgo": "sha256-normalized-v1",
-  "generatedAt": "2026-10-10T00:13:36.699Z",
+  "generatedAt": "2026-10-10T13:32:39.437Z",
   "project": {
     "name": "AgentsKit",
     "root": "."
@@ -3687,7 +3687,7 @@
         "document",
         "human"
       ],
-      "contentHash": "ce0b77e00540b9f6d2004aca322bd6497f39b505d0df62d23ee42266ff3e5334"
+      "contentHash": "2f8e5695adf8368b45b9e2fff381159ba18324b21e29ae8777b459030f194353"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/angular.mdx",
@@ -3699,7 +3699,7 @@
         "document",
         "human"
       ],
-      "contentHash": "722ff97e8f3a9245c8d8b23d2074d418c85e5eed5ee0f6f2a2bbf30a6bf8930f"
+      "contentHash": "5b901eb6c16394298beb5a209936add3cc1bf48928965876621d7efb66b51372"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/cli.mdx",
@@ -3711,7 +3711,7 @@
         "document",
         "human"
       ],
-      "contentHash": "5b0f721867b15dadb92bc276877ccb476f5fb793da0a76ec934f23a9140d5ee8"
+      "contentHash": "89b93eb9ebacbd746030f3f819fd82f986946bd64cd54c8786ec4f94c24a6cc8"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/core.mdx",
@@ -3723,7 +3723,7 @@
         "document",
         "human"
       ],
-      "contentHash": "c0805db69400cae3a7a0abc8f1dbacfc9bae0078798f4a2157df89c22c5bdf32"
+      "contentHash": "4a39bd5e7491df761748fa1f3056725b798b531b9029a278e5eb2c521ab9a204"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/cross-platform.mdx",
@@ -3735,19 +3735,19 @@
         "document",
         "human"
       ],
-      "contentHash": "374df5fb9fe1af9d8511f5f078ea51fae5e2d590eefc531f3c5f4a888131ec76"
+      "contentHash": "7d26dbe12d5015d730d985034ed1fafcfb09cf692da825ca1cf69e5b9b968bad"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx",
       "type": "document",
       "title": "eval-braintrust.mdx",
       "path": "apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx",
-      "description": "Version: 0.2.27",
+      "description": "Version: 0.2.28",
       "tags": [
         "document",
         "human"
       ],
-      "contentHash": "5c4b5678f866df934b9ff6517b32b13d369ac082365c3a3be9f2fddd6c852178"
+      "contentHash": "c27e7532aa7ce86050a4a81094463d0c28312d2ad57ad537627dd3c49fa70454"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/eval.mdx",
@@ -3759,7 +3759,7 @@
         "document",
         "human"
       ],
-      "contentHash": "5b094f44d37ae1636b18487709905e175260c763623c4e9ee43cdb77333bfe6e"
+      "contentHash": "d423029df433bd2bc94ae01398baa92d75cb81049ae7ec7b977be5a39fa045e5"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/ink.mdx",
@@ -3771,7 +3771,7 @@
         "document",
         "human"
       ],
-      "contentHash": "1987e6bdd4c5e711858a7fb9b7eb2a19bcdb379b788edcbb0a5d558e8ea0d2c5"
+      "contentHash": "73cc8ffc847ffb58525a333c01a204d5eb3154b3009aceeec2ace65156f6cca4"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/memory.mdx",
@@ -3783,7 +3783,7 @@
         "document",
         "human"
       ],
-      "contentHash": "c9dcb00062dea5c73961e35b846df51a8143f83336b3a1b5df95262bcd83894a"
+      "contentHash": "8dd9f05860de712e131eb3c96eae190716c20212f9c3595023836729b5eb90bd"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/net.mdx",
@@ -3795,19 +3795,19 @@
         "document",
         "human"
       ],
-      "contentHash": "042fd6d2df5eac857182760e88aed72b2491ca17c32b1be4ee6587395bb4c610"
+      "contentHash": "c5c16474c2949034eb3ba0c226abf9dd5661633975de783a245b81e3d271b2b9"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx",
       "type": "document",
       "title": "observability-langfuse.mdx",
       "path": "apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx",
-      "description": "Version: 0.2.26",
+      "description": "Version: 0.2.27",
       "tags": [
         "document",
         "human"
       ],
-      "contentHash": "c49ede22b1482583096c04ce6c7ab8862a3b7c6bf2d6914a608c891df418e439"
+      "contentHash": "acb7a721d0c3fea44e57fa429d9778edd8e7312a82b37e5b79599a180416bdb0"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/observability.mdx",
@@ -3819,7 +3819,7 @@
         "document",
         "human"
       ],
-      "contentHash": "ef1b809976de1c818cb8a99839f929d4a9fa5bf13b634598684abffad563e6aa"
+      "contentHash": "4daf7c1c8db1aafdf83d90c55dee29c94d5172669877f29c66dc241df1aafa0f"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/overview.mdx",
@@ -3843,7 +3843,7 @@
         "document",
         "human"
       ],
-      "contentHash": "c76ee7a3326f5d0fc84ed374abf4e6d79b4649dd329eb731c722a00e9455df5c"
+      "contentHash": "ef350a62dd4e53a8a700209289ece498cad7ec12adec467e9c47da3d86611a7f"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/react-native.mdx",
@@ -3855,7 +3855,7 @@
         "document",
         "human"
       ],
-      "contentHash": "63c0330c2ac7f272a142a6892869095135793c29a24032d64321b85109a4f6b4"
+      "contentHash": "d87cb4cb70523fe8788862d0292bf5fd464b546bae84a186558d2b3799ef86c0"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/react.mdx",
@@ -3867,7 +3867,7 @@
         "document",
         "human"
       ],
-      "contentHash": "21f699615d3ee6117b7677f145804fbcb2a4d026d2882f5417648415850ec615"
+      "contentHash": "a4bb6bf390e371c106562e75f09cdd7cf038954689170a7a843db6e275874be2"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/roadmap.mdx",
@@ -3891,7 +3891,7 @@
         "document",
         "human"
       ],
-      "contentHash": "acb56d36d6ab24f5fd464105ff92e0823db0ea6d39a94e7b3ce670cce8adb9c5"
+      "contentHash": "e036a7f1a0e0d59200f2d478a727021591159b71b63373ae9fbd2072dfb5dbef"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/sandbox.mdx",
@@ -3903,7 +3903,7 @@
         "document",
         "human"
       ],
-      "contentHash": "71a06422fcac73ee88ae3d23750bb1009d6025a0397f8fbd1ebcfbe1381d0944"
+      "contentHash": "12e59255242a51d64b5606cf98a4337af41a3b6d23b4ffa97e284f1084fc25a5"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/skills.mdx",
@@ -3915,7 +3915,7 @@
         "document",
         "human"
       ],
-      "contentHash": "b9aaa3a1ba16d15fc211e7bd644727228d57701427bf0e0d1cba49a6aec0d318"
+      "contentHash": "e31dbc239e29ad8ae34513b06423ce5f047265ae5eaef29511fa1fc407dfa3cc"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/solid.mdx",
@@ -3927,7 +3927,7 @@
         "document",
         "human"
       ],
-      "contentHash": "98b1da134eb7bbc88694b281a148cca5746ffe86a2f3259cfad296c7cdb82f15"
+      "contentHash": "2e10d7d9889e291f39aec531fd3a8dee2da718097cd6c49d4b5c30439deb409b"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/statechart.mdx",
@@ -3939,7 +3939,7 @@
         "document",
         "human"
       ],
-      "contentHash": "9a18c08d7f9891e5d25c5d8831a7f6f3f6cd9a65e2463b381f8bee5166932541"
+      "contentHash": "7c6b9fdf86438681027bdb62ed5b1bb88be3d06a2d2e6db203eb62360ddef1e0"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/svelte.mdx",
@@ -3951,7 +3951,7 @@
         "document",
         "human"
       ],
-      "contentHash": "372bfce01f26f0589d9417bc067fd53b277d2e2c854655b09039bd672d77486b"
+      "contentHash": "7bb3adde2a6b7058f26494178710cd4d0f30d075e942da1e8363238b20ccf05b"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/templates.mdx",
@@ -3963,7 +3963,7 @@
         "document",
         "human"
       ],
-      "contentHash": "ab7be64918967558608f217cd6d69b6f461ca84adc094f69d164b98c5cc155be"
+      "contentHash": "10706eeb53c2aa2f2ff85ef946531adc81acb2aad00b92750f27ca72c373ca6e"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/tools.mdx",
@@ -3975,7 +3975,7 @@
         "document",
         "human"
       ],
-      "contentHash": "1926701dc379e3eaee8b94876ffff67e2a2fac797adef09f1e35265d6e8b05b2"
+      "contentHash": "f191d6d99e58f14104dcb644021c464391703a59bdc726c7f47f02681d60ada2"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/vue.mdx",
@@ -3987,7 +3987,7 @@
         "document",
         "human"
       ],
-      "contentHash": "1d37433f2377c39748a6ee6a5d68755f70974f23aedefc6488a1b20159973ec4"
+      "contentHash": "57a94acb35de8805ad81d8ba9e1eb18b62c529495c0e8edd7e378aba31f7de2e"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/publications.mdx",
@@ -4131,7 +4131,7 @@
         "document",
         "human"
       ],
-      "contentHash": "7fbecf065621c8074f25b15d6589aa68628998d04950e879b6ef71709a2dc1e0"
+      "contentHash": "6c0deafff5744176d351b6c42aae19daab54983b5e6f1ee286ec8d0547ca21b5"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/recipes/confirmation-gated-tool.mdx",
@@ -4395,7 +4395,7 @@
         "document",
         "human"
       ],
-      "contentHash": "a81ab301cdde13f04a8b4ae4d85328ef1e54c65083b5f12624a8e58dc5c54648"
+      "contentHash": "78128080568b626ffc2fa51aecb47bd4117340555df3724668407d9727ad5123"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/recipes/more-integrations.mdx",
@@ -6074,7 +6074,7 @@
         "document",
         "human"
       ],
-      "contentHash": "9f0c1e6420206c1f6a813589cd231af59b72d22fb68a895a6bafed1d6f1b8203"
+      "contentHash": "6bbcb59353c45545b97728da92df345307d0642d32582c46c731d181393edf64"
     },
     {
       "id": "document:docs/architecture/adrs/README.md",
@@ -6146,7 +6146,7 @@
         "document",
         "human"
       ],
-      "contentHash": "2c3613d9b17a0ab1ae8076e55a0f885968b06b566cc34d446b7d9094b58988e7"
+      "contentHash": "98a26068fa0636dd6d79762b1b3606616e5b823f49ee4f86718f147446fdf8fd"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md",
@@ -6158,7 +6158,7 @@
         "document",
         "human"
       ],
-      "contentHash": "e13286472557d766283be88550b5db9fc0ee583c4521ab5a236b7848e5399dfb"
+      "contentHash": "90e92591b6bc6b4b991729c7eda9deb21e37bbe5ff482f40c713691e82438c4d"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx",
@@ -6170,7 +6170,7 @@
         "document",
         "human"
       ],
-      "contentHash": "2a89209aae9627e8fc934630b0fb27e7861283fbb558f9bfd97c661128ef725b"
+      "contentHash": "0c786965cab639ef1728feb7b4355039740f699811aaac3bd529a33e9ea89fc6"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/review.md",
@@ -6206,7 +6206,7 @@
         "document",
         "human"
       ],
-      "contentHash": "8807e7a4ace7ed7bb5f507aa82b588068ebcbb97cc12aba5331aa56100c32f6f"
+      "contentHash": "05e7837ff75b4b53c65053d8843d6d575ba35bcc12d4ee157c5d5d04cd6b9310"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/first-agent/carousel-storyboard.md",
@@ -7190,7 +7190,7 @@
         "document",
         "unclassified"
       ],
-      "contentHash": "d140f0cf8621abad4dab454537ebab797c96004bec9da916c579828703254ff5"
+      "contentHash": "4306e774390c1c87a40c6cb508eab760bdb0886c2d0b5ac5ce0180e90ba9c574"
     },
     {
       "id": "document:packages/mcp/README.md",
@@ -7202,7 +7202,7 @@
         "document",
         "project"
       ],
-      "contentHash": "44223f4f7f15d48dc74085432f785211877a9a1a84ab09f2946ae9fac44aacb5"
+      "contentHash": "3097341299c6313b1a1b6cfafddb2149faf3137477da3d953f6a9495dc8cd8f7"
     },
     {
       "id": "document:packages/memory/CONVENTIONS.md",
@@ -7298,7 +7298,7 @@
         "document",
         "project"
       ],
-      "contentHash": "f784f385ea347b6b418c6d84d67f7c6a100a21806c880640f9edbfdcf1635221"
+      "contentHash": "87237cd2b57ba5c3b22587d59167010673075706fdfde7ff9b2eb344783f4de1"
     },
     {
       "id": "document:packages/rag/CONVENTIONS.md",
@@ -25722,7 +25722,7 @@
         "ts",
         "fixtures"
       ],
-      "contentHash": "e937255484065fc80d6a9d5508b896caf651356180b91ddd8142c59d6b8db3bb"
+      "contentHash": "999745c52fe403d21ecd122d746b8f48f0c519ed557c4f9a58d19a52341d729f"
     },
     {
       "id": "module:packages/mcp/fixtures/readme-quickstart.ts",
@@ -25758,7 +25758,7 @@
         "mjs",
         "fixtures"
       ],
-      "contentHash": "13a202839963acc9607784c70d35d9d2cc1d2fcde7c32e170708d97729201fc6"
+      "contentHash": "7d404d08513159d609e6039a3ed49c4a0a0294d9d629515d51d3dbdcadb989d7"
     },
     {
       "id": "module:packages/mcp/src/agent-tool.ts",
@@ -25902,7 +25902,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "76f2f46976359b2d942e774336963692bffb0c5865fb52845db70bfa91e3b063"
+      "contentHash": "090e00a1a759cff1439b2606e7571e19d27cd6f92f9d20284d712caeb8f7f363"
     },
     {
       "id": "module:packages/mcp/tests/package-manifest.test.ts",
@@ -27934,7 +27934,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "75f302e345d2a4bd69fbdf08b4693eedaa50a59c2c6a019fcc1774913d5ed12f"
+      "contentHash": "362f6a41a929494e30cb3dcba902a6d67c15ef163fe9e83d7524706528d6308c"
     },
     {
       "id": "module:packages/observability/src/cost-guard-alert-sinks.ts",
@@ -27968,7 +27968,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "6577cedb8e48993548baefaae96fbc910d157fe3e5d355345ff3d286c0fa5dc3"
+      "contentHash": "ebf4e6e4938bc3f0ea7a9791cf6a6168b97d45380428630d1885007b90ddb989"
     },
     {
       "id": "module:packages/observability/src/cost-guard-store.ts",
@@ -27986,7 +27986,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "9a078a80a71859f4200812350e14376f90fd49f0b602bed5e006e4054c18259b"
+      "contentHash": "f790b633d6eaff1bc10da4fd92388afcf8dd51ae9cd2222c434409a11cba9aad"
     },
     {
       "id": "module:packages/observability/src/cost-guard.ts",
@@ -28018,7 +28018,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "dbed63fe3d0288ce8719ea849709b60870774e988be501b23fe69a731785355d"
+      "contentHash": "7a574710105c8a6d41a5f3615e20ca713dd8db041620b1fcefa2648ead6e089d"
     },
     {
       "id": "module:packages/observability/src/cost-store-contract.ts",
@@ -28035,7 +28035,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "1e5f787b952cbbd9bdfd9b809cbdf7a8c315ee9a3def164285420482c228f189"
+      "contentHash": "f523e7509cb4e0f9c8901d9ad0dcf071142527b03e4d139b693e4b647e0598b5"
     },
     {
       "id": "module:packages/observability/src/cost-store.ts",
@@ -28061,7 +28061,9 @@
         "monthlyWindowKey",
         "recordCostSpend",
         "validateCostCommit",
+        "validateCostRelease",
         "validateCostReserve",
+        "validateCostWindow",
         "warnProcessLocalCostState"
       ],
       "tags": [
@@ -28069,7 +28071,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "000d000f32f8ff879d6372400a7620b9574392e10947972ffa252c1339d51dfe"
+      "contentHash": "a4de0beaaa030a3cccabedb0b68e616d9333997bde4ca6ce6f2251203b2bae81"
     },
     {
       "id": "module:packages/observability/src/datadog.ts",
@@ -28338,7 +28340,9 @@
       "title": "postgres.ts",
       "path": "packages/observability/src/postgres.ts",
       "symbols": [
+        "ExpirePostgresCostReservationsOptions",
         "PostgresCostStoreOptions",
+        "expirePostgresCostReservations",
         "postgresCostMigrationSql",
         "postgresCostReservationTable",
         "postgresCostStore",
@@ -28350,7 +28354,7 @@
         "ts",
         "src"
       ],
-      "contentHash": "c717f504c6644d0f7d1db9017415638bc8f1889e8e974545c508aa31484da167"
+      "contentHash": "08672acc3b8820e6a96bdd2e30521771091138ce0a5484ab1749e5d931bbbe23"
     },
     {
       "id": "module:packages/observability/src/prod-control.ts",
@@ -28555,6 +28559,18 @@
       ],
       "contentHash": "eed9b27f639570ceaeadca73cdf916339d0a3cf60613fe5b1de579d3fb98ee1a"
     },
+    {
+      "id": "module:packages/observability/tests/cjs-smoke.cjs",
+      "type": "module",
+      "title": "cjs-smoke.cjs",
+      "path": "packages/observability/tests/cjs-smoke.cjs",
+      "tags": [
+        "module",
+        "cjs",
+        "tests"
+      ],
+      "contentHash": "56ca4f72a7dbbd0a9eafecc08c5c3678b1e767896657fee8d5525b1c4e814427"
+    },
     {
       "id": "module:packages/observability/tests/console-logger.test.ts",
       "type": "module",
@@ -28657,7 +28673,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "3dbca16e7462eb3d4946f12a3730c0396d4a0762cc414572d1e6d820e0d74288"
+      "contentHash": "f0bd5884a15e5771ce295772ddca61703b9d9f4b7edf47188486a18d7d254e98"
     },
     {
       "id": "module:packages/observability/tests/cost-guard.test.ts",
@@ -28683,7 +28699,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "447ac4782c545122cb71dfc17c03751951c9e62ec542b47d91d2df3e2768e3f8"
+      "contentHash": "3a284757bbc12d9d061ea11791f9b4e8e9cb9b9a69e4cf78b630f88aaa4f9008"
     },
     {
       "id": "module:packages/observability/tests/cost-store.test.ts",
@@ -28696,7 +28712,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "263a1943013345075c1080f4b31228f7fe934bc31feb05d2f0b5c0ab9fd42f1d"
+      "contentHash": "3a7ffc7e3fca98ac3da8ed711ee91f5c57e4d24b8d744f4e87c24d286ba6c9a6"
     },
     {
       "id": "module:packages/observability/tests/devtools.test.ts",
@@ -28777,7 +28793,7 @@
         "ts",
         "tests"
       ],
-      "contentHash": "f3d3e7a3eae7a1c1d5cddb2bf506ea81fdc0fff40056edbd57505a02e1b4611b"
+      "contentHash": "5950721a855c0536a6d54b89b013d507fb567b63df59a2cbbd669a04b3ce5289"
     },
     {
       "id": "module:packages/observability/tests/postgres-worker.ts",
@@ -28792,7 +28808,7 @@
         "ts",
         "tests"
       ],
-      "contentHash": "407e64dc7114efd7920d36a27a747899ccb99afbf79db8a86c2ec404df756f12"
+      "contentHash": "9e78a9f4cc0757bbe13b5f4e8f149d62c4f8d6df9e6e82678e1f5d3c34a1428b"
     },
     {
       "id": "module:packages/observability/tests/postgres-workerd.mjs",
@@ -28804,7 +28820,7 @@
         "mjs",
         "tests"
       ],
-      "contentHash": "7a7a6a85faffe15c65b44898f95505d629d8aa57a2d7c3750c0a187a15f92bce"
+      "contentHash": "dbd4f1b1f4c518fdc519e586f9b2655a6856d036d084504108b47fc410b72a69"
     },
     {
       "id": "module:packages/observability/tests/postgres.integration.test.ts",
@@ -28817,7 +28833,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "8fb55eaf341dfedbc79f7ce382f60e785b3954afb810e36048803a02c7c1feb7"
+      "contentHash": "301ce5fb51c3d0910a5eba60c067ab36ee65c26c879b2882cef099bb6fed0156"
     },
     {
       "id": "module:packages/observability/tests/postgres.test.ts",
@@ -28830,7 +28846,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "9f3255e0a36e2632e7119af79ca31ee09386584d3b1f1d6e9df665710ac3f76a"
+      "contentHash": "21268cb747a51e617a2679d73362734f99c6f338e9766d7e72f97f4d805aba0f"
     },
     {
       "id": "module:packages/observability/tests/prod-control.test.ts",
@@ -38572,8 +38588,8 @@
     "changes": {}
   },
   "inputs": {
-    "hash": "6fed057dcbc0aba6f59e7843774e3abaffcd6c3bb1df390b6059e1ea22ec6abf",
-    "fileCount": 2562,
+    "hash": "52545bde66c9523c0506d4fbffc30570141194f7fd0bae765f438864f272dd09",
+    "fileCount": 2563,
     "projectionVersion": 2
   },
   "retrieval": {
@@ -38595,9 +38611,9 @@
   "projection": {
     "type": "retrieval-index",
     "schemaVersion": 1,
-    "contentHash": "a79c78312df2218ccf1c4fb17ecd51b3c64eafb5eb0b7afd9de693c88931d169",
+    "contentHash": "cf62905beac13114d8dc6dfbabad6c726df7117aba0e7df14ca5733b87b85a82",
     "contentHashAlgo": "sha256-normalized-v1",
-    "snapshotHash": "2c2951311fc560f749cec14351cb1b672e5c0acccadbd4ef880b2d69997e2dd9",
+    "snapshotHash": "c4df5c892dcfa99507e733f574eaeff92af12de91742aafda7d28cd0fd986a98",
     "overlayHash": "f99666ff9119084dd25c2974ae64eb6935ec2aa785ff5ad478fb9d541f7a5e94",
     "configurationHash": "9662cc07ea4ab203ff2ddf457d2db48b5375a7ab4a1f8fbc79dfebcaa8f9a00c",
     "lexiconVersion": 1,
@@ -38617,7 +38633,7 @@
     },
     "lexical": {
       "version": 1,
-      "documentCount": 2517,
+      "documentCount": 2518,
       "fieldNames": [
         "aliases",
         "body",
@@ -38628,13 +38644,13 @@
         "title"
       ],
       "averageFieldLength": {
-        "aliases": 3.0154946364719906,
-        "body": 61.76042908224076,
-        "headings": 4.018275725069527,
-        "path": 7.612236789829161,
-        "summary": 4.54906634882797,
-        "symbols": 7.5013905442987685,
-        "title": 3.3663090981326977
+        "aliases": 3.015488482922955,
+        "body": 61.772041302621126,
+        "headings": 4.016679904686259,
+        "path": 7.611993645750596,
+        "summary": 4.5472597299444,
+        "symbols": 7.505957108816521,
+        "title": 3.3665607625099283
       }
     },
     "entries": [
@@ -52258,7 +52274,7 @@
           "path": "apps/docs-next/content/docs/for-agents/mcp.mdx",
           "symbols": "",
           "summary": "Serve AgentsKit tools to an MCP host that supports the documented `2024-11-05` tools bridge. A thin bridge over `@agentskit/tools/mcp` (`createMcpServer` + `createStdioTransport`).",
-          "body": "Purpose Serve AgentsKit tools to an MCP host that supports the documented 2024-11-05 tools bridge. A thin bridge over @agentskit/tools/mcp ( createMcpServer + createStdioTransport ). Install Primary exports - createAgentsKitMcpServer({ tools, serverInfo?, onEvent?, transport?, authorizeToolCall? }) — expose McpToolDefinition[] as an MCP server. Defaults to stdio over the current process; pass transport to override (tests / custom hosts). - processStdio() — adapt the Node process into the StdioLikeProcess the transport expects (server receives on stdin, sends on stdout). - createAgentTool({ id, description, systemPrompt, adapter, maxSteps? }) — wrap a whole agent as one MCP tool. The host calls it with a task ; the agent runs server-side on the given adapter and returns the result (\"agents as MCP tools\"). - createTypedAgentTool(config) — typed variant of createAgentTool for an agent input/output contract. - fetchAgent(id, fetchImpl?) — fetch a registry agent definition without executing its source. - fetchAgentSkill(id, fetchImpl?) — fetch a registry agent's runnable skill (hosted index, raw-GitHub fallback); null for tool-composing agents. An optional third argument controls abort, timeout, and response-byte limits. Pass authorizeToolCall to createAgentsKitMcpServer to obtain host approval for tools marked requiresConfirmation . It has the same type as McpServerOptions['authorizeToolCall'] : (tool, args) = boolean Promise<boolean . Arguments are validated first. Only confirmation-required tools invoke the hook; absent hooks, denied approvals, throws, and rejected promises prevent execution. The host must obtain the required human confirmation before returning true ; annotations do not replace that decision. McpToolDefinition (from @agentskit/tools/mcp ) extends the core tool shape with optional title and annotations : title , readOnlyHint , destructiveHint , idempotentHint , and openWorldHint . Server listing and client import preserve these fields, including explicit false hints. Imported titles use the existing maxDescriptionBytes UTF-8 limit; malformed metadata is rejected. Hints remain untrusted advisory data and never grant confirmation or alter execution policy. These fields follow the . They are optional descriptor extensions on the existing 2024-11-05 bridge; older clients may ignore unknown fields. This does not add negotiation or other features from newer protocol revisions. Bin agentskit-mcp — npx -y @agentskit/mcp@0.4.10 --tools fetch,search . Expose whole agents with --agents <ids --provider <p (any of ~20 providers + catalog OpenAI-compatible). Flags: --tools , --fs-root , --sqlite , --allow-shell , --model , --max-steps , --api-key , --base-url . Unknown, duplicate, or valueless flags fail closed. stdout is the JSON-RPC channel; diagnostics and help go to stderr. Pinned coding-agent configs and their evidence levels live in the . Host config fixtures cover Codex, Claude Code/Desktop, Cursor, Cline, and Continue. The local-build and published-package smoke scripts both initialize the shared STDIO path and require the tool list to equal fetch url,web search . Beta boundaries - Tool names are unique and follow the current MCP 1–128 character interoperability grammar. - Server configuration and agent-tool configuration fail with typed AgentsKit diagnostics. - Published tool definitions are snapshotted; observer failures cannot alter protocol behavior. - Agent tasks, step counts, prompts, descriptions, registry IDs, request time, and response bytes are bounded. - Registry source is treated as data and never executed. - The built-in surface is the MCP tools subset over stdio or an injected transport. HTTP/WebSocket, resources, prompts, sampling, auth, rate limiting, and persistence are host responsibilities. Supported protocol matrix Protocol revision Lifecycle Transports Methods Not included --- --- --- --- --- 2024-11-05 Exact-version initialize ; tools/list ; tools/call ; close settles pending calls stdio, in-memor",
+          "body": "Purpose Serve AgentsKit tools to an MCP host that supports the documented 2024-11-05 tools bridge. A thin bridge over @agentskit/tools/mcp ( createMcpServer + createStdioTransport ). Install Primary exports - createAgentsKitMcpServer({ tools, serverInfo?, onEvent?, transport?, authorizeToolCall? }) — expose McpToolDefinition[] as an MCP server. Defaults to stdio over the current process; pass transport to override (tests / custom hosts). - processStdio() — adapt the Node process into the StdioLikeProcess the transport expects (server receives on stdin, sends on stdout). - createAgentTool({ id, description, systemPrompt, adapter, maxSteps? }) — wrap a whole agent as one MCP tool. The host calls it with a task ; the agent runs server-side on the given adapter and returns the result (\"agents as MCP tools\"). - createTypedAgentTool(config) — typed variant of createAgentTool for an agent input/output contract. - fetchAgent(id, fetchImpl?) — fetch a registry agent definition without executing its source. - fetchAgentSkill(id, fetchImpl?) — fetch a registry agent's runnable skill (hosted index, raw-GitHub fallback); null for tool-composing agents. An optional third argument controls abort, timeout, and response-byte limits. Pass authorizeToolCall to createAgentsKitMcpServer to obtain host approval for tools marked requiresConfirmation . It has the same type as McpServerOptions['authorizeToolCall'] : (tool, args) = boolean Promise<boolean . Arguments are validated first. Only confirmation-required tools invoke the hook; absent hooks, denied approvals, throws, and rejected promises prevent execution. The host must obtain the required human confirmation before returning true ; annotations do not replace that decision. McpToolDefinition (from @agentskit/tools/mcp ) extends the core tool shape with optional title and annotations : title , readOnlyHint , destructiveHint , idempotentHint , and openWorldHint . Server listing and client import preserve these fields, including explicit false hints. Imported titles use the existing maxDescriptionBytes UTF-8 limit; malformed metadata is rejected. Hints remain untrusted advisory data and never grant confirmation or alter execution policy. These fields follow the . They are optional descriptor extensions on the existing 2024-11-05 bridge; older clients may ignore unknown fields. This does not add negotiation or other features from newer protocol revisions. Bin agentskit-mcp — npx -y @agentskit/mcp@0.5.0 --tools fetch,search . Expose whole agents with --agents <ids --provider <p (any of ~20 providers + catalog OpenAI-compatible). Flags: --tools , --fs-root , --sqlite , --allow-shell , --model , --max-steps , --api-key , --base-url . Unknown, duplicate, or valueless flags fail closed. stdout is the JSON-RPC channel; diagnostics and help go to stderr. Pinned coding-agent configs and their evidence levels live in the . Host config fixtures cover Codex, Claude Code/Desktop, Cursor, Cline, and Continue. The local-build and published-package smoke scripts both initialize the shared STDIO path and require the tool list to equal fetch url,web search . Beta boundaries - Tool names are unique and follow the current MCP 1–128 character interoperability grammar. - Server configuration and agent-tool configuration fail with typed AgentsKit diagnostics. - Published tool definitions are snapshotted; observer failures cannot alter protocol behavior. - Agent tasks, step counts, prompts, descriptions, registry IDs, request time, and response bytes are bounded. - Registry source is treated as data and never executed. - The built-in surface is the MCP tools subset over stdio or an injected transport. HTTP/WebSocket, resources, prompts, sampling, auth, rate limiting, and persistence are host responsibilities. Supported protocol matrix Protocol revision Lifecycle Transports Methods Not included --- --- --- --- --- 2024-11-05 Exact-version initialize ; tools/list ; tools/call ; close settles pending calls stdio, in-memory",
           "aliases": "mcp document agent"
         },
         "graph": {
@@ -52320,7 +52336,7 @@
             }
           ]
         },
-        "contentHash": "b435f20208aa75706de84a879b0e9ad4e78f1ce732203c859114ac28605b9899",
+        "contentHash": "ff7095d5548807583af0d9a1211cb01ff9e4104e4fa2adb99eceda934c988c87",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -52642,7 +52658,7 @@
           "path": "apps/docs-next/content/docs/for-agents/observability.mdx",
           "symbols": "",
           "summary": "consoleLogger(config?) — local dev logging.",
-          "body": "Install Primary exports Loggers / tracers - consoleLogger(config?) — local dev logging. - langsmith(config) — LangSmith lifecycle observer ( flush / idempotent shutdown ; optional langsmith peer, lazy). - opentelemetry(config) — OpenTelemetry lifecycle observer (optional OTel peers, lazy). - createTraceTracker({ onSpanStart, onSpanEnd }) — low-level span lifecycle. Cross-repository correlation Every AgentEvent may carry an optional correlation envelope. Set the stable operationId at the host boundary and preserve local runId , sessionId , turnId , actionId , and traceId values when they exist. Observability propagates these values as span attributes without requiring a vendor SDK or a shared runtime. Sinks (batch HTTP exporters) All three return LifecycleObserver ( flush(): Promise<void , idempotent shutdown(): Promise<void ). Shared HttpBatchOptions defaults: batchSize 25, maxQueueSize 1000 (drop-oldest), flushIntervalMs 2000, maxRetries 3, retryBaseDelayMs 100, requestTimeoutMs 10000, plus isolated onError . Single-flight batching, timeout, and retries are best-effort — not an at-least-once guarantee. Await shutdown() on graceful termination. - datadogSink(config) — Datadog Logs intake. - axiomSink(config) — Axiom dataset ingest. - newRelicSink(config) — New Relic Log API. Vendor adapters - @agentskit/observability/langfuse — Langfuse logger backend. Install the optional langfuse peer when using this subpath. Guards + counters - costGuard({ budgetUsd, controller, prices?, unknownModelPolicy?, onCost?, onExceeded?, onError? }) — per-run dollar ceiling; aborts via AbortController when exceeded and fails closed on unknown pricing by default. - multiTenantCostGuard({ budgets, defaultBudgetUsd?, tenantOf?, unknownModelPolicy?, … }) — per-tenant bookkeeping; host enforces (no automatic abort), and unknown pricing marks the tenant blocked by default. - createAdvancedCostGuard({ budgets, caps?, mode?, unknownModelPolicy?, disableRuntime?, alertSinks?, onError?, now? }) — production guard with rolling window caps, 50/80/100 % threshold + forecast alerts, and modes: - warn — observe only. - reject — host must consult isRejected(tenant) (window rejections clear on roll; overall until reset ). Package does not abort the runtime. - kill — requires disableRuntime ; use isDisabled(tenant) ; fail-closed if disable fails. - Accounting is incremental per llm:end / active model ; hostile usage normalizes to zero; zero-budget payloads stay finite. Callback/sink/clock errors are isolated. - consoleAlertSink() , webhookAlertSink({ url, headers, fetch? }) , throttle(sink, windowMs) — built-in alert sinks ( webhook uses injected or globalThis.fetch ; no-op if none). - store?: CostStore on all three guards — durable spend shared by every instance. Without it totals are process-local and the guards warn once when NODE ENV=production . With it, each priced llm:end is written to the store and the budget is checked against the store's current window. - CostStore { reserve, commit, release, window } — tenant-scoped port. reserve({ tenant, reservationId, amountUsd, capUsd?, windowKey? }) returns { ok: true, window } or { ok: false, reason: 'quota exceeded', window } ; commit({ tenant, reservationId, actualUsd, usage? }) writes real spend and ledger rows; release({ tenant, reservationId }) returns the hold. All idempotent per reservationId ; settling an unknown, released or committed reservation throws CostReservationError ( AK COST RESERVATION NOT FOUND / RELEASED / COMMITTED ). - createInMemoryCostStore() — development and test store. recordCostSpend(store, { tenant, costUsd, usage? }) — reserve + commit for spend that already happened. monthlyWindowKey(date) — default window ( YYYY-MM , UTC). - @agentskit/observability/postgres — postgresCostStore({ db }) , postgresCostMigrationSql , and the Drizzle tables postgresCostWindowTable , postgresCostReservationTable , postgresUsageLedgerTable . One statement per operation, cap admitted atomically. Optional pe",
+          "body": "Install Primary exports Loggers / tracers - consoleLogger(config?) — local dev logging. - langsmith(config) — LangSmith lifecycle observer ( flush / idempotent shutdown ; optional langsmith peer, lazy). - opentelemetry(config) — OpenTelemetry lifecycle observer (optional OTel peers, lazy). - createTraceTracker({ onSpanStart, onSpanEnd }) — low-level span lifecycle. Cross-repository correlation Every AgentEvent may carry an optional correlation envelope. Set the stable operationId at the host boundary and preserve local runId , sessionId , turnId , actionId , and traceId values when they exist. Observability propagates these values as span attributes without requiring a vendor SDK or a shared runtime. Sinks (batch HTTP exporters) All three return LifecycleObserver ( flush(): Promise<void , idempotent shutdown(): Promise<void ). Shared HttpBatchOptions defaults: batchSize 25, maxQueueSize 1000 (drop-oldest), flushIntervalMs 2000, maxRetries 3, retryBaseDelayMs 100, requestTimeoutMs 10000, plus isolated onError . Single-flight batching, timeout, and retries are best-effort — not an at-least-once guarantee. Await shutdown() on graceful termination. - datadogSink(config) — Datadog Logs intake. - axiomSink(config) — Axiom dataset ingest. - newRelicSink(config) — New Relic Log API. Vendor adapters - @agentskit/observability/langfuse — Langfuse logger backend. Install the optional langfuse peer when using this subpath. Guards + counters - costGuard({ budgetUsd, controller, prices?, unknownModelPolicy?, onCost?, onExceeded?, onError? }) — per-run dollar ceiling; aborts via AbortController when exceeded and fails closed on unknown pricing by default. - multiTenantCostGuard({ budgets, defaultBudgetUsd?, tenantOf?, unknownModelPolicy?, … }) — per-tenant bookkeeping; host enforces (no automatic abort), and unknown pricing marks the tenant blocked by default. - createAdvancedCostGuard({ budgets, caps?, mode?, unknownModelPolicy?, disableRuntime?, alertSinks?, onError?, now? }) — production guard with rolling window caps, 50/80/100 % threshold + forecast alerts, and modes: - warn — observe only. - reject — host must consult isRejected(tenant) (window rejections clear on roll; overall until reset ). Package does not abort the runtime. - kill — requires disableRuntime ; use isDisabled(tenant) ; fail-closed if disable fails. - Accounting is incremental per llm:end / active model ; hostile usage normalizes to zero; zero-budget payloads stay finite. Callback/sink/clock errors are isolated. - consoleAlertSink() , webhookAlertSink({ url, headers, fetch? }) , throttle(sink, windowMs) — built-in alert sinks ( webhook uses injected or globalThis.fetch ; no-op if none). - store?: CostStore on all three guards — durable spend shared by every instance. Without it totals are process-local and the guards warn once when NODE ENV=production (or on Cloudflare Workers with NODE ENV unset); warnWithoutStore?: boolean overrides that. Store failures go to onError , or to one console.warn when it is absent. With it, each priced llm:end is written to the store and the budget is checked against the store's current window. - CostStore { reserve, commit, release, window } — tenant-scoped port. reserve({ tenant, reservationId, amountUsd, capUsd?, windowKey? }) returns { ok: true, window } or { ok: false, reason: 'quota exceeded', window } ; commit({ tenant, reservationId, actualUsd, usage? }) writes real spend and ledger rows; release({ tenant, reservationId }) returns the hold. All idempotent per reservationId ; settling an unknown, released or committed reservation throws CostReservationError ( AK COST RESERVATION NOT FOUND / RELEASED / COMMITTED ). - createInMemoryCostStore() — development and test store. recordCostSpend(store, { tenant, costUsd, usage? }) — reserve + commit for spend that already happened. monthlyWindowKey(date) — default window ( YYYY-MM , UTC). - @agentskit/observability/postgres — postgresCostStore({ db }) , postgresCostMigrationSql , and the Dr",
           "aliases": "observability document agent"
         },
         "graph": {
@@ -52759,7 +52775,7 @@
             }
           ]
         },
-        "contentHash": "238366434038e8d05ba7d34619655cd867c0b30c34d91cf6c57a8422cdc42008",
+        "contentHash": "4527b56550623200e878944b69d9e156ecf8a0fae6abb07a2f5d871aeb511e98",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -59929,7 +59945,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/adapters.mdx",
           "symbols": "",
           "summary": "@agentskit/adapters is usually the first package you touch after core. It is the provider seam for the entire ecosystem: UI, runtime, memory, evals, and recipes all plug into this layer.",
-          "body": "@agentskit/adapters is usually the first package you touch after core . It is the provider seam for the entire ecosystem: UI, runtime, memory, evals, and recipes all plug into this layer. When to reach for it - You need to talk to a hosted LLM (Anthropic, OpenAI, Gemini, Mistral, Cohere, Groq, Together, Fireworks, OpenRouter, Hugging Face, …). - You want local-only (Ollama, LM Studio, vLLM, llama.cpp). - You want to compose multiple candidates ( createRouter , createEnsembleAdapter , createFallbackAdapter ). - You want to test agents without hitting a real LLM ( mockAdapter , recordingAdapter , replayAdapter ). Best fit - Start here if your main concern is provider flexibility. - Pair with if you are shipping a chat product. - Pair with if the model needs to act over multiple steps. - Pair with if you want to compare providers over the same workload. Install Hello world The important part is not the constructor itself, but that the rest of your stack does not have to care which provider you chose. Catalog The @agentskit/adapters/catalog subpath: data-driven provider/model metadata adapted from into AgentsKit's own JSON Schema and cached as a committed snapshot. Broad, current coverage lands via a version bump instead of bespoke per-provider code. The snapshot is large (thousands of models), so it ships only via the ./catalog subpath — the main @agentskit/adapters bundle is unaffected. Provider and model counts are regenerated from the artifact rather than hand-maintained. No runtime fetch. The runtime loads the committed snapshot; it never calls models.dev . If models.dev ever disappears, the last snapshot keeps working. Refresh it with pnpm sync:models ; the snapshot includes a normalized content hash and upstream ETag. A scheduled workflow compares the committed hash with the normalized source and opens a draft refresh PR when the catalog changes; it also fails when the snapshot is older than 35 days: The snapshot carries generatedAt + a pinned source.version ( catalogSource() ) so you can reason about staleness. Pricing/limits are advisory cached metadata , not a hard contract. Pricing with optional live fallback. resolveCost() is cache-only by default (offline, deterministic). Opt in to a live lookup that falls back to the cached snapshot on any failure — it never throws on a network problem: Policy overrides ( applyOverrides ) constrain the catalog without forking it (allowed/disabled providers, per-provider model allow-lists). Drift ( detectCatalogDrift ) flags any snapshot provider that is neither first-class nor OpenAI-compatible — wire it into CI so a regenerated snapshot can't ship an unroutable provider silently. classifyCatalogProvider(provider) exposes the explicit native / openai-compatible / unsupported matrix so catalog breadth is never confused with transport coverage. The catalog schema ( catalogSnapshotSchema , JSON Schema) is the public contract; the snapshot is validated against it at build time. Surface Hosted: anthropic · openai · gemini · grok · deepseek · kimi · mistral · cohere · together · groq · fireworks · openrouter · huggingface · langchain · langgraph · vercelAI · generic Local: ollama · lmstudio · vllm · llamacpp Embedders: openaiEmbedder · geminiEmbedder · ollamaEmbedder · deepseekEmbedder · grokEmbedder · kimiEmbedder · createOpenAICompatibleEmbedder Higher-order: createRouter · createEnsembleAdapter · createFallbackAdapter Testing: mockAdapter · recordingAdapter · replayAdapter · inMemorySink · simulateStream · chunkText · fetchWithRetry Recipes - - - - - - Stability - Version: 0.18.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/adapters is usually the first package you touch after core . It is the provider seam for the entire ecosystem: UI, runtime, memory, evals, and recipes all plug into this layer. When to reach for it - You need to talk to a hosted LLM (Anthropic, OpenAI, Gemini, Mistral, Cohere, Groq, Together, Fireworks, OpenRouter, Hugging Face, …). - You want local-only (Ollama, LM Studio, vLLM, llama.cpp). - You want to compose multiple candidates ( createRouter , createEnsembleAdapter , createFallbackAdapter ). - You want to test agents without hitting a real LLM ( mockAdapter , recordingAdapter , replayAdapter ). Best fit - Start here if your main concern is provider flexibility. - Pair with if you are shipping a chat product. - Pair with if the model needs to act over multiple steps. - Pair with if you want to compare providers over the same workload. Install Hello world The important part is not the constructor itself, but that the rest of your stack does not have to care which provider you chose. Catalog The @agentskit/adapters/catalog subpath: data-driven provider/model metadata adapted from into AgentsKit's own JSON Schema and cached as a committed snapshot. Broad, current coverage lands via a version bump instead of bespoke per-provider code. The snapshot is large (thousands of models), so it ships only via the ./catalog subpath — the main @agentskit/adapters bundle is unaffected. Provider and model counts are regenerated from the artifact rather than hand-maintained. No runtime fetch. The runtime loads the committed snapshot; it never calls models.dev . If models.dev ever disappears, the last snapshot keeps working. Refresh it with pnpm sync:models ; the snapshot includes a normalized content hash and upstream ETag. A scheduled workflow compares the committed hash with the normalized source and opens a draft refresh PR when the catalog changes; it also fails when the snapshot is older than 35 days: The snapshot carries generatedAt + a pinned source.version ( catalogSource() ) so you can reason about staleness. Pricing/limits are advisory cached metadata , not a hard contract. Pricing with optional live fallback. resolveCost() is cache-only by default (offline, deterministic). Opt in to a live lookup that falls back to the cached snapshot on any failure — it never throws on a network problem: Policy overrides ( applyOverrides ) constrain the catalog without forking it (allowed/disabled providers, per-provider model allow-lists). Drift ( detectCatalogDrift ) flags any snapshot provider that is neither first-class nor OpenAI-compatible — wire it into CI so a regenerated snapshot can't ship an unroutable provider silently. classifyCatalogProvider(provider) exposes the explicit native / openai-compatible / unsupported matrix so catalog breadth is never confused with transport coverage. The catalog schema ( catalogSnapshotSchema , JSON Schema) is the public contract; the snapshot is validated against it at build time. Surface Hosted: anthropic · openai · gemini · grok · deepseek · kimi · mistral · cohere · together · groq · fireworks · openrouter · huggingface · langchain · langgraph · vercelAI · generic Local: ollama · lmstudio · vllm · llamacpp Embedders: openaiEmbedder · geminiEmbedder · ollamaEmbedder · deepseekEmbedder · grokEmbedder · kimiEmbedder · createOpenAICompatibleEmbedder Higher-order: createRouter · createEnsembleAdapter · createFallbackAdapter Testing: mockAdapter · recordingAdapter · replayAdapter · inMemorySink · simulateStream · chunkText · fetchWithRetry Recipes - - - - - - Stability - Version: 0.19.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60146,7 +60162,7 @@
             }
           ]
         },
-        "contentHash": "ce0b77e00540b9f6d2004aca322bd6497f39b505d0df62d23ee42266ff3e5334",
+        "contentHash": "2f8e5695adf8368b45b9e2fff381159ba18324b21e29ae8777b459030f194353",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60168,7 +60184,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/angular.mdx",
           "symbols": "",
           "summary": "AgentskitChat — @Injectable({ providedIn: 'root' }):",
-          "body": "Install Hello world Surface - AgentskitChat — @Injectable({ providedIn: 'root' }) : - init(config) — bootstrap the controller. - state: WritableSignal<ChatState — template-friendly. - stream$: Observable<ChatState — RxJS. - Actions: send · stop · retry · setInput · clear · approve · deny . Siblings · · · · · Stability - Version: 0.5.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
+          "body": "Install Hello world Surface - AgentskitChat — @Injectable({ providedIn: 'root' }) : - init(config) — bootstrap the controller. - state: WritableSignal<ChatState — template-friendly. - stream$: Observable<ChatState — RxJS. - Actions: send · stop · retry · setInput · clear · approve · deny . Siblings · · · · · Stability - Version: 0.5.10 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60205,7 +60221,7 @@
             }
           ]
         },
-        "contentHash": "722ff97e8f3a9245c8d8b23d2074d418c85e5eed5ee0f6f2a2bbf30a6bf8930f",
+        "contentHash": "5b901eb6c16394298beb5a209936add3cc1bf48928965876621d7efb66b51372",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60227,7 +60243,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/cli.mdx",
           "symbols": "",
           "summary": "@agentskit/cli is the operational front door to the ecosystem. It is how you bootstrap a project, run an agent, inspect your setup, and keep moving without writing scaffolding code first.",
-          "body": "@agentskit/cli is the operational front door to the ecosystem. It is how you bootstrap a project, run an agent, inspect your setup, and keep moving without writing scaffolding code first. When to reach for it - You want to scaffold a new agent project. - You want to chat / run an agent from a terminal. - You want agentskit ai \"<description \" to generate a typed scaffold from plain language. Best fit - Start here when you want proof of the ecosystem before wiring packages by hand. - Use init when you want a runnable starter. - Use chat , run , and dev when you want a fast operator workflow around the same stack. - Use doctor when setup friction is slowing the team down. Install Commands Command Purpose --- --- agentskit init Scaffold a new project (react / ink / runtime / multi-agent). agentskit chat Interactive chat in the terminal (Ink). agentskit run \"<task \" Run an agent once. agentskit dev Dev server with hot-reload. agentskit doctor Diagnose env (providers, keys, tooling). agentskit ai \"<description \" NL → typed AgentSchema + scaffolded project. agentskit tunnel ngrok-style tunnel for webhooks. agentskit rag Local RAG helpers (ingest / search). agentskit config Read / write local config. Why it matters The CLI makes AgentsKit feel like a product, not just a pile of packages. It shortens the path from “I want to try this” to “I have a working agent project” more than any single reference page can. Programmatic helpers - @agentskit/cli/ai — scaffoldAgent · writeScaffold · createAdapterPlanner . Recipes - Stability - Version: 0.13.46 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/cli is the operational front door to the ecosystem. It is how you bootstrap a project, run an agent, inspect your setup, and keep moving without writing scaffolding code first. When to reach for it - You want to scaffold a new agent project. - You want to chat / run an agent from a terminal. - You want agentskit ai \"<description \" to generate a typed scaffold from plain language. Best fit - Start here when you want proof of the ecosystem before wiring packages by hand. - Use init when you want a runnable starter. - Use chat , run , and dev when you want a fast operator workflow around the same stack. - Use doctor when setup friction is slowing the team down. Install Commands Command Purpose --- --- agentskit init Scaffold a new project (react / ink / runtime / multi-agent). agentskit chat Interactive chat in the terminal (Ink). agentskit run \"<task \" Run an agent once. agentskit dev Dev server with hot-reload. agentskit doctor Diagnose env (providers, keys, tooling). agentskit ai \"<description \" NL → typed AgentSchema + scaffolded project. agentskit tunnel ngrok-style tunnel for webhooks. agentskit rag Local RAG helpers (ingest / search). agentskit config Read / write local config. Why it matters The CLI makes AgentsKit feel like a product, not just a pile of packages. It shortens the path from “I want to try this” to “I have a working agent project” more than any single reference page can. Programmatic helpers - @agentskit/cli/ai — scaffoldAgent · writeScaffold · createAdapterPlanner . Recipes - Stability - Version: 0.13.47 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60274,7 +60290,7 @@
             }
           ]
         },
-        "contentHash": "5b0f721867b15dadb92bc276877ccb476f5fb793da0a76ec934f23a9140d5ee8",
+        "contentHash": "89b93eb9ebacbd746030f3f819fd82f986946bd64cd54c8786ec4f94c24a6cc8",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60438,7 +60454,7 @@
             }
           ]
         },
-        "contentHash": "c0805db69400cae3a7a0abc8f1dbacfc9bae0078798f4a2157df89c22c5bdf32",
+        "contentHash": "4a39bd5e7491df761748fa1f3056725b798b531b9029a278e5eb2c521ab9a204",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60460,7 +60476,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/cross-platform.mdx",
           "symbols": "",
           "summary": "@agentskit/cross-platform is the package for the moment your agent (or your tooling) has to run a command, compare a path or parse a file and it must work for the teammate on Windows as well as on your Mac and in Linux CI — whether the code runs on Node, Bun or Deno.",
-          "body": "@agentskit/cross-platform is the package for the moment your agent (or your tooling) has to run a command, compare a path or parse a file and it must work for the teammate on Windows as well as on your Mac and in Linux CI — whether the code runs on Node, Bun or Deno. When to reach for it - You spawn CLIs ( npx , pnpm , claude , codex , git ) and need them to start on Windows without shell: true . - You pass long or multi-line prompts to a child process. - You need a timeout that kills the whole process tree, not only the wrapper. - You compare, store or hash paths and text produced on different machines. Best fit - Use it instead of node:child process in any AgentsKit package or ecosystem repository. - Run agentskit-cross-platform check in CI so new Windows hazards fail on Linux, before a Windows user finds them. - Use @agentskit/cross-platform/pure in browsers and edge workers (paths, text, runtime detection). Install Hello world What it replaces Hand-rolled pattern Use --- --- spawn(cmd, args, { shell: process.platform === 'win32' }) spawnProcess(cmd, args) Prompt passed as an argument runCommand(cmd, args, { input }) child.kill() / process.kill(-pid) timeoutMs , handle.kill() , killProcessTree(pid) p.replace(/\\\\/g, '/') toPosix(p) / relativePosix(a, b) new URL('.', import.meta.url).pathname moduleDir(import.meta.url) text.split('\\n') splitLines(text) rename/rm retry loops renamePath , removePath , writeFileAtomic Testing helpers @agentskit/cross-platform/testing exports framework-agnostic helpers for shared test setup: createFetchStub routes URL requests to a handler or response map, jsonResponse creates a JSON Response , and withTempDir cleans up its temporary directory after the callback completes. Runtimes One native adapter per runtime: node:child process on Node, Bun.spawn on Bun, Deno.Command on Deno. stdout and stderr are web ReadableStream s everywhere. On Deno, a missing --allow- flag raises AK PLATFORM PERMISSION DENIED naming the flag. Stability - Version: 0.2.1 - Tier: beta - Contract: evolving Deep dives - — full export list. - Decision record: ADR-0036 ( docs/architecture/adrs/0036-cross-platform-package.md ).",
+          "body": "@agentskit/cross-platform is the package for the moment your agent (or your tooling) has to run a command, compare a path or parse a file and it must work for the teammate on Windows as well as on your Mac and in Linux CI — whether the code runs on Node, Bun or Deno. When to reach for it - You spawn CLIs ( npx , pnpm , claude , codex , git ) and need them to start on Windows without shell: true . - You pass long or multi-line prompts to a child process. - You need a timeout that kills the whole process tree, not only the wrapper. - You compare, store or hash paths and text produced on different machines. Best fit - Use it instead of node:child process in any AgentsKit package or ecosystem repository. - Run agentskit-cross-platform check in CI so new Windows hazards fail on Linux, before a Windows user finds them. - Use @agentskit/cross-platform/pure in browsers and edge workers (paths, text, runtime detection). Install Hello world What it replaces Hand-rolled pattern Use --- --- spawn(cmd, args, { shell: process.platform === 'win32' }) spawnProcess(cmd, args) Prompt passed as an argument runCommand(cmd, args, { input }) child.kill() / process.kill(-pid) timeoutMs , handle.kill() , killProcessTree(pid) p.replace(/\\\\/g, '/') toPosix(p) / relativePosix(a, b) new URL('.', import.meta.url).pathname moduleDir(import.meta.url) text.split('\\n') splitLines(text) rename/rm retry loops renamePath , removePath , writeFileAtomic Testing helpers @agentskit/cross-platform/testing exports framework-agnostic helpers for shared test setup: createFetchStub routes URL requests to a handler or response map, jsonResponse creates a JSON Response , and withTempDir cleans up its temporary directory after the callback completes. Runtimes One native adapter per runtime: node:child process on Node, Bun.spawn on Bun, Deno.Command on Deno. stdout and stderr are web ReadableStream s everywhere. On Deno, a missing --allow- flag raises AK PLATFORM PERMISSION DENIED naming the flag. Stability - Version: 0.3.0 - Tier: beta - Contract: evolving Deep dives - — full export list. - Decision record: ADR-0036 ( docs/architecture/adrs/0036-cross-platform-package.md ).",
           "aliases": "document human"
         },
         "graph": {
@@ -60497,7 +60513,7 @@
             }
           ]
         },
-        "contentHash": "374df5fb9fe1af9d8511f5f078ea51fae5e2d590eefc531f3c5f4a888131ec76",
+        "contentHash": "7d26dbe12d5015d730d985034ed1fafcfb09cf692da825ca1cf69e5b9b968bad",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60506,7 +60522,7 @@
         "kind": "document",
         "path": "apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx",
         "title": "eval-braintrust.mdx",
-        "summary": "Version: 0.2.27",
+        "summary": "Version: 0.2.28",
         "audience": "human",
         "aliases": [],
         "tags": [
@@ -60518,8 +60534,8 @@
           "headings": "When to reach for it Install Public API Root (@agentskit/eval/braintrust) Scorers (@agentskit/eval/braintrust/scorers) CI helpers (@agentskit/eval/braintrust/ci) Minimal example CI regression check Configuration Stability Related Source",
           "path": "apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx",
           "symbols": "",
-          "summary": "Version: 0.2.27",
-          "body": "- Version: 0.2.27 @agentskit/eval/braintrust connects AgentsKit eval suites to . It runs your test cases through deterministic scorers, optionally ships explicitly selected results to a Braintrust experiment, and surfaces regressions as CI annotations — all without requiring Braintrust to be installed at build time (peer-resolved at runtime). The implementation is a private workspace source package; consumers install the public @agentskit/eval package and its /braintrust subpath. Version: 0.2.19 When to reach for it - You already use Braintrust for experiment tracking and want AgentsKit runs to show up in its experiment UI. - You want multi-dimensional scoring (task success, factual grounding, citation correctness, schema survival, HITL gate, crash resilience) with a single import. - You want regression alerts between two experiment snapshots in CI. Install braintrust is an optional peer — the adapter loads it dynamically at runtime. If the key / package is absent, results are scored locally and no remote sync occurs. Public API Root ( @agentskit/eval/braintrust ) Export Kind Purpose --- --- --- runBraintrustEval(args, internals?) async fn Run cases, score, optionally sync to Braintrust, return ExperimentResult scoreCase(scorers, input) async fn Score a single ScorerInput against an array of scorers summarize(cases) fn Aggregate ScoredCase[] into per-scorer { mean, n } map BraintrustRunOptions type Options passed to runBraintrustEval ScoredCase type Single case result with scores + duration ExperimentResult type Full run result: cases, summary, remote URL RunBraintrustEvalArgs type Argument shape for runBraintrustEval Scorer type (args: ScorerInput) = ScorerResult \\ Promise<ScorerResult ScorerInput type { input, output, expected?, metadata? } ScorerResult type { name, score, rationale?, metadata? } ScorerFamily type { family: 'quality' \\ 'robustness'; scorers } Scorers ( @agentskit/eval/braintrust/scorers ) Export Family What it measures --- --- --- taskSuccess quality Did the output satisfy the task? factualGrounding quality Is the output grounded in provided context? citationCorrectness quality Are citations present and accurate? toolArgValidity quality Are tool call arguments schema-valid? schemaSurvival robustness Does output survive schema round-trip? hitlGateCorrectness robustness Did HITL gate fire correctly? fallbackResilience robustness Does agent recover from injected failures? noCrashSurvival robustness Does agent complete without throwing? qualityFamily — All four quality scorers bundled robustnessFamily — All four robustness scorers bundled ALL SCORERS — Flat array of all eight scorers CI helpers ( @agentskit/eval/braintrust/ci ) Export Purpose --- --- detectRegressions(baseline, current, thresholds?) Returns RegressionAlert[] for scorers that dropped beyond threshold formatAlertsMarkdown(alerts) Renders a markdown table of regressions for PR comments / CI summary RegressionThresholds { default?: number; perScorer?: Record<string, number } RegressionAlert { scorer, baseline, current, delta, threshold } Minimal example CI regression check Configuration Env var Purpose --- --- BRAINTRUST API KEY Explicitly enables Braintrust upload. Without it, scoring runs locally and no data is uploaded. BRAINTRUST BASE URL Override the Braintrust API base URL (default: Braintrust cloud). You can also pass apiKey / baseUrl directly in BraintrustRunOptions — explicit values take precedence over env vars. Remote case fields are excluded by default. To upload selected fields, set options.upload explicitly with includeInput , includeOutput , includeExpected , or an allowlist of metadataKeys ; every selected value is capped by maxFieldBytes (default 4096). Run-level options.metadata follows the same allowlist and is never sent by default. Stability - Version: inherited from @agentskit/eval ; the workspace adapter source is not published independently. - Tier: beta - Contract: scorer shapes stable; Braintrust SDK peer-resolved at runtim",
+          "summary": "Version: 0.2.28",
+          "body": "- Version: 0.2.28 @agentskit/eval/braintrust connects AgentsKit eval suites to . It runs your test cases through deterministic scorers, optionally ships explicitly selected results to a Braintrust experiment, and surfaces regressions as CI annotations — all without requiring Braintrust to be installed at build time (peer-resolved at runtime). The implementation is a private workspace source package; consumers install the public @agentskit/eval package and its /braintrust subpath. Version: 0.2.19 When to reach for it - You already use Braintrust for experiment tracking and want AgentsKit runs to show up in its experiment UI. - You want multi-dimensional scoring (task success, factual grounding, citation correctness, schema survival, HITL gate, crash resilience) with a single import. - You want regression alerts between two experiment snapshots in CI. Install braintrust is an optional peer — the adapter loads it dynamically at runtime. If the key / package is absent, results are scored locally and no remote sync occurs. Public API Root ( @agentskit/eval/braintrust ) Export Kind Purpose --- --- --- runBraintrustEval(args, internals?) async fn Run cases, score, optionally sync to Braintrust, return ExperimentResult scoreCase(scorers, input) async fn Score a single ScorerInput against an array of scorers summarize(cases) fn Aggregate ScoredCase[] into per-scorer { mean, n } map BraintrustRunOptions type Options passed to runBraintrustEval ScoredCase type Single case result with scores + duration ExperimentResult type Full run result: cases, summary, remote URL RunBraintrustEvalArgs type Argument shape for runBraintrustEval Scorer type (args: ScorerInput) = ScorerResult \\ Promise<ScorerResult ScorerInput type { input, output, expected?, metadata? } ScorerResult type { name, score, rationale?, metadata? } ScorerFamily type { family: 'quality' \\ 'robustness'; scorers } Scorers ( @agentskit/eval/braintrust/scorers ) Export Family What it measures --- --- --- taskSuccess quality Did the output satisfy the task? factualGrounding quality Is the output grounded in provided context? citationCorrectness quality Are citations present and accurate? toolArgValidity quality Are tool call arguments schema-valid? schemaSurvival robustness Does output survive schema round-trip? hitlGateCorrectness robustness Did HITL gate fire correctly? fallbackResilience robustness Does agent recover from injected failures? noCrashSurvival robustness Does agent complete without throwing? qualityFamily — All four quality scorers bundled robustnessFamily — All four robustness scorers bundled ALL SCORERS — Flat array of all eight scorers CI helpers ( @agentskit/eval/braintrust/ci ) Export Purpose --- --- detectRegressions(baseline, current, thresholds?) Returns RegressionAlert[] for scorers that dropped beyond threshold formatAlertsMarkdown(alerts) Renders a markdown table of regressions for PR comments / CI summary RegressionThresholds { default?: number; perScorer?: Record<string, number } RegressionAlert { scorer, baseline, current, delta, threshold } Minimal example CI regression check Configuration Env var Purpose --- --- BRAINTRUST API KEY Explicitly enables Braintrust upload. Without it, scoring runs locally and no data is uploaded. BRAINTRUST BASE URL Override the Braintrust API base URL (default: Braintrust cloud). You can also pass apiKey / baseUrl directly in BraintrustRunOptions — explicit values take precedence over env vars. Remote case fields are excluded by default. To upload selected fields, set options.upload explicitly with includeInput , includeOutput , includeExpected , or an allowlist of metadataKeys ; every selected value is capped by maxFieldBytes (default 4096). Run-level options.metadata follows the same allowlist and is never sent by default. Stability - Version: inherited from @agentskit/eval ; the workspace adapter source is not published independently. - Tier: beta - Contract: scorer shapes stable; Braintrust SDK peer-resolved at runtim",
           "aliases": "document human"
         },
         "graph": {
@@ -60581,7 +60597,7 @@
             }
           ]
         },
-        "contentHash": "5c4b5678f866df934b9ff6517b32b13d369ac082365c3a3be9f2fddd6c852178",
+        "contentHash": "c27e7532aa7ce86050a4a81094463d0c28312d2ad57ad537627dd3c49fa70454",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60603,7 +60619,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/eval.mdx",
           "symbols": "",
           "summary": "@agentskit/eval is how you stop shipping changes to prompts, tools, and providers on vibes alone. It gives you repeatable ways to measure quality, compare behavior, and catch regressions before they reach users.",
-          "body": "@agentskit/eval is how you stop shipping changes to prompts, tools, and providers on vibes alone. It gives you repeatable ways to measure quality, compare behavior, and catch regressions before they reach users. When to reach for it - You want to score agent quality with numbers, in CI. - You want deterministic replay (record once, replay forever). - You want Jest-style prompt snapshots with semantic tolerance. - You want a \"git blame for prompts\" — diff + attribution. Best fit - Add this when an agent starts becoming product-critical. - Pair with so you can turn real failures into eval cases. - Pair with to compare providers against the same suite. - Pair with when you need to evaluate full multi-step workflows, not isolated prompts. Install Hello world That feedback loop is what lets a team keep improving an agent without losing control of it. Surface - runEval({ agent, suite }) . - /provenance : universal createResultRecord · serializeResultRecord · verifyResultRecord · precisionRecall . - /replay : universal in-memory createRecordingAdapter · createReplayAdapter · cassettes · createTimeTravelSession · replayAgainst · summarizeReplay . - /replay/io : Node-only saveCassette · loadCassette . Browser and native applications should persist serialized cassettes through a host-owned storage adapter. - /snapshot : matchPromptSnapshot . - /diff : promptDiff · attributePromptChange · formatDiff . - /ci : renderJUnit · renderMarkdown · renderGitHubAnnotations · reportToCi . Result records Use @agentskit/eval/provenance to preserve benchmark identity and metrics without running an agent: ResultRecordInput requires non-empty suiteId , subject.revision , and runnerVersion , lowercase 64-character SHA-256 caseSetDigest and subject.digest , and a JSON metrics object ( ResultMetric supports nested arrays and objects). Creation snapshots the input. ResultRecord adds digest : SHA-256 of the UTF-8 RFC 8785 canonical JSON of all input fields except timestamp . Serialization includes the digest and optional timestamp. Omit the timestamp for byte-identical reruns; changing it does not invalidate the digest. Core's canonical JSON semantics apply: undefined object fields are omitted, undefined array entries become null, and non-finite numbers and cycles are rejected. verifyResultRecord(unknown) returns false for malformed provenance or a mismatched digest, without throwing. It checks integrity, not authenticity or whether the supplied revision and digests correspond to real artifacts. A party able to rewrite the record and its digest can produce another valid record. precisionRecall({ tp, fp, fn }) accepts non-negative safe integers and returns { precision, recall } in [0, 1] . A zero denominator yields zero. Counts and per-unit results can be stored directly in metrics; these helpers do not run or judge cases. Recipes - - - - - - - Stability - Version: 0.6.12 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/eval is how you stop shipping changes to prompts, tools, and providers on vibes alone. It gives you repeatable ways to measure quality, compare behavior, and catch regressions before they reach users. When to reach for it - You want to score agent quality with numbers, in CI. - You want deterministic replay (record once, replay forever). - You want Jest-style prompt snapshots with semantic tolerance. - You want a \"git blame for prompts\" — diff + attribution. Best fit - Add this when an agent starts becoming product-critical. - Pair with so you can turn real failures into eval cases. - Pair with to compare providers against the same suite. - Pair with when you need to evaluate full multi-step workflows, not isolated prompts. Install Hello world That feedback loop is what lets a team keep improving an agent without losing control of it. Surface - runEval({ agent, suite }) . - /provenance : universal createResultRecord · serializeResultRecord · verifyResultRecord · precisionRecall . - /replay : universal in-memory createRecordingAdapter · createReplayAdapter · cassettes · createTimeTravelSession · replayAgainst · summarizeReplay . - /replay/io : Node-only saveCassette · loadCassette . Browser and native applications should persist serialized cassettes through a host-owned storage adapter. - /snapshot : matchPromptSnapshot . - /diff : promptDiff · attributePromptChange · formatDiff . - /ci : renderJUnit · renderMarkdown · renderGitHubAnnotations · reportToCi . Result records Use @agentskit/eval/provenance to preserve benchmark identity and metrics without running an agent: ResultRecordInput requires non-empty suiteId , subject.revision , and runnerVersion , lowercase 64-character SHA-256 caseSetDigest and subject.digest , and a JSON metrics object ( ResultMetric supports nested arrays and objects). Creation snapshots the input. ResultRecord adds digest : SHA-256 of the UTF-8 RFC 8785 canonical JSON of all input fields except timestamp . Serialization includes the digest and optional timestamp. Omit the timestamp for byte-identical reruns; changing it does not invalidate the digest. Core's canonical JSON semantics apply: undefined object fields are omitted, undefined array entries become null, and non-finite numbers and cycles are rejected. verifyResultRecord(unknown) returns false for malformed provenance or a mismatched digest, without throwing. It checks integrity, not authenticity or whether the supplied revision and digests correspond to real artifacts. A party able to rewrite the record and its digest can produce another valid record. precisionRecall({ tp, fp, fn }) accepts non-negative safe integers and returns { precision, recall } in [0, 1] . A zero denominator yields zero. Counts and per-unit results can be stored directly in metrics; these helpers do not run or judge cases. Recipes - - - - - - - Stability - Version: 0.7.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60690,7 +60706,7 @@
             }
           ]
         },
-        "contentHash": "5b094f44d37ae1636b18487709905e175260c763623c4e9ee43cdb77333bfe6e",
+        "contentHash": "d423029df433bd2bc94ae01398baa92d75cb81049ae7ec7b977be5a39fa045e5",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60712,7 +60728,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/ink.mdx",
           "symbols": "",
           "summary": "You need a CLI / terminal chat.",
-          "body": "When to reach for it - You need a CLI / terminal chat. - You want to embed an agent inside a dev tool. Install Hello world Surface - useChat(config): ChatReturn — shares the core state/action contract; Ink components and terminal input semantics are host-specific. - <ChatContainer , <Message , <InputBar , <ToolCallView , <ThinkingIndicator . Siblings · · · · · Stability - Version: 0.10.15 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - (ships agentskit chat powered by Ink) Source npm: · repo:",
+          "body": "When to reach for it - You need a CLI / terminal chat. - You want to embed an agent inside a dev tool. Install Hello world Surface - useChat(config): ChatReturn — shares the core state/action contract; Ink components and terminal input semantics are host-specific. - <ChatContainer , <Message , <InputBar , <ToolCallView , <ThinkingIndicator . Siblings · · · · · Stability - Version: 0.10.16 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - (ships agentskit chat powered by Ink) Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60734,7 +60750,7 @@
             }
           ]
         },
-        "contentHash": "1987e6bdd4c5e711858a7fb9b7eb2a19bcdb379b788edcbb0a5d558e8ea0d2c5",
+        "contentHash": "73cc8ffc847ffb58525a333c01a204d5eb3154b3009aceeec2ace65156f6cca4",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60756,7 +60772,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/memory.mdx",
           "symbols": "",
           "summary": "@agentskit/memory is the layer you add when your agent needs continuity. Without it, every run starts cold. With it, the system can remember conversations, preferences, retrieved context, and higher-order abstractions over time.",
-          "body": "@agentskit/memory is the layer you add when your agent needs continuity. Without it, every run starts cold. With it, the system can remember conversations, preferences, retrieved context, and higher-order abstractions over time. When to reach for it - You need persistent chat history (Web Storage / file / SQLite / Redis). - You need vector search (pgvector / Pinecone / Qdrant / Chroma / Upstash / Redis / file). - You need MemGPT-style tiered memory, or client-side encryption, or a knowledge graph. Best fit - Start here when the same user, workflow, or corpus needs to persist across sessions. - Pair with for agent continuity. - Pair with for retrieval-heavy assistants. - Pair with if you need visibility into memory behavior and quality. Install Hello world In most real products, memory is the difference between a clever demo and a useful system. Surface - Chat: createWebStorageMemory · fileChatMemory · sqliteChatMemory · redisChatMemory . - Vector: fileVectorMemory · redisVectorMemory · pgvector · pinecone · qdrant · chroma · upstashVector . - HoF wrappers: createHierarchicalMemory · createEncryptedMemory · createInMemoryGraph · createInMemoryPersonalization . Recipes - - - - - - - Stability - Version: 0.11.14 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/memory is the layer you add when your agent needs continuity. Without it, every run starts cold. With it, the system can remember conversations, preferences, retrieved context, and higher-order abstractions over time. When to reach for it - You need persistent chat history (Web Storage / file / SQLite / Redis). - You need vector search (pgvector / Pinecone / Qdrant / Chroma / Upstash / Redis / file). - You need MemGPT-style tiered memory, or client-side encryption, or a knowledge graph. Best fit - Start here when the same user, workflow, or corpus needs to persist across sessions. - Pair with for agent continuity. - Pair with for retrieval-heavy assistants. - Pair with if you need visibility into memory behavior and quality. Install Hello world In most real products, memory is the difference between a clever demo and a useful system. Surface - Chat: createWebStorageMemory · fileChatMemory · sqliteChatMemory · redisChatMemory . - Vector: fileVectorMemory · redisVectorMemory · pgvector · pinecone · qdrant · chroma · upstashVector . - HoF wrappers: createHierarchicalMemory · createEncryptedMemory · createInMemoryGraph · createInMemoryPersonalization . Recipes - - - - - - - Stability - Version: 0.12.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60868,7 +60884,7 @@
             }
           ]
         },
-        "contentHash": "c9dcb00062dea5c73961e35b846df51a8143f83336b3a1b5df95262bcd83894a",
+        "contentHash": "8dd9f05860de712e131eb3c96eae190716c20212f9c3595023836729b5eb90bd",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60890,7 +60906,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/net.mdx",
           "symbols": "",
           "summary": "@agentskit/net is the package for the moment your agent talks HTTP: a provider returns 429 with Retry-After, a stream arrives as server-sent events, a tool fetches a URL a user typed, or a response could be larger than you want in memory. It uses web APIs only, so the same code runs on Node, Bun, Deno, browsers and edge runtimes.",
-          "body": "@agentskit/net is the package for the moment your agent talks HTTP: a provider returns 429 with Retry-After , a stream arrives as server-sent events, a tool fetches a URL a user typed, or a response could be larger than you want in memory. It uses web APIs only, so the same code runs on Node, Bun, Deno, browsers and edge runtimes. When to reach for it - You call an API that rate-limits or fails transiently and need retries that respect Retry-After . - You need a timeout per attempt, not one that silently spans every retry. - You need a deadline for async work and a signal that forwards caller cancellation. - You consume a server-sent events stream. - You fetch URLs supplied by users or models and must not reach private networks or cloud metadata. Install Hello world Cancellable waits and deadlines retry uses the exported sleep(ms, signal?) by default. Delays must be from 0 through 2,147,483,647 milliseconds. It clears its timer on abort and rejects with the signal's exact reason. To replace the wait, pass sleep: (ms, signal?) = Promise<void ; one-argument callbacks still work, and callbacks should honor the optional signal when their own wait must stop on cancellation. withTimeout(work, ms, parent?) accepts a positive deadline up to 2,147,483,647 milliseconds and gives work a signal that aborts at the deadline or with the parent's exact reason. The deadline rejects with AK NET TIMEOUT ; invalid values reject with AK NET INVALID INPUT . It bounds the wrapper's rejection even if work ignores the signal, but only cooperative work such as fetch can be stopped. A success value or original work failure passes through unchanged. retry retries rejected work, such as fetch network errors; it does not inspect HTTP status codes. Use fetchWithRetry when retrying retryable HTTP responses. What it replaces Hand-rolled pattern Use --- --- for loop with a fixed sleep retry(fn) / fetchWithRetry(url) Ignoring or re-parsing Retry-After <code parseRetryAfter<wbr / (header)</code (built into fetchWithRetry ) Promise.race with setTimeout timeoutSignal(ms) , anySignal([...]) , withTimeout(work, ms) await res.text() on unknown input readText(res, { maxBytes }) line.startsWith('data: ') parseSSE(stream) Regexes over 127. , 10. , 192.168. <code assertPublicUrl<wbr / (url, & 123; lookup & 125;)</code Stability - Version: 0.2.2 - Tier: beta - Contract: evolving Deep dives - — full export list. - Decision record: ADR-0037 ( docs/architecture/adrs/0037-net-package.md ).",
+          "body": "@agentskit/net is the package for the moment your agent talks HTTP: a provider returns 429 with Retry-After , a stream arrives as server-sent events, a tool fetches a URL a user typed, or a response could be larger than you want in memory. It uses web APIs only, so the same code runs on Node, Bun, Deno, browsers and edge runtimes. When to reach for it - You call an API that rate-limits or fails transiently and need retries that respect Retry-After . - You need a timeout per attempt, not one that silently spans every retry. - You need a deadline for async work and a signal that forwards caller cancellation. - You consume a server-sent events stream. - You fetch URLs supplied by users or models and must not reach private networks or cloud metadata. Install Hello world Cancellable waits and deadlines retry uses the exported sleep(ms, signal?) by default. Delays must be from 0 through 2,147,483,647 milliseconds. It clears its timer on abort and rejects with the signal's exact reason. To replace the wait, pass sleep: (ms, signal?) = Promise<void ; one-argument callbacks still work, and callbacks should honor the optional signal when their own wait must stop on cancellation. withTimeout(work, ms, parent?) accepts a positive deadline up to 2,147,483,647 milliseconds and gives work a signal that aborts at the deadline or with the parent's exact reason. The deadline rejects with AK NET TIMEOUT ; invalid values reject with AK NET INVALID INPUT . It bounds the wrapper's rejection even if work ignores the signal, but only cooperative work such as fetch can be stopped. A success value or original work failure passes through unchanged. retry retries rejected work, such as fetch network errors; it does not inspect HTTP status codes. Use fetchWithRetry when retrying retryable HTTP responses. What it replaces Hand-rolled pattern Use --- --- for loop with a fixed sleep retry(fn) / fetchWithRetry(url) Ignoring or re-parsing Retry-After <code parseRetryAfter<wbr / (header)</code (built into fetchWithRetry ) Promise.race with setTimeout timeoutSignal(ms) , anySignal([...]) , withTimeout(work, ms) await res.text() on unknown input readText(res, { maxBytes }) line.startsWith('data: ') parseSSE(stream) Regexes over 127. , 10. , 192.168. <code assertPublicUrl<wbr / (url, & 123; lookup & 125;)</code Stability - Version: 0.2.3 - Tier: beta - Contract: evolving Deep dives - — full export list. - Decision record: ADR-0037 ( docs/architecture/adrs/0037-net-package.md ).",
           "aliases": "document human"
         },
         "graph": {
@@ -60922,7 +60938,7 @@
             }
           ]
         },
-        "contentHash": "042fd6d2df5eac857182760e88aed72b2491ca17c32b1be4ee6587395bb4c610",
+        "contentHash": "c5c16474c2949034eb3ba0c226abf9dd5661633975de783a245b81e3d271b2b9",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60931,7 +60947,7 @@
         "kind": "document",
         "path": "apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx",
         "title": "observability-langfuse.mdx",
-        "summary": "Version: 0.2.26",
+        "summary": "Version: 0.2.27",
         "audience": "human",
         "aliases": [],
         "tags": [
@@ -60943,8 +60959,8 @@
           "headings": "When to reach for it Best fit Install Hello world Public API langfuse(config?): Observer LangfuseConfig Per-session tracing Self-hosted Langfuse Configuration Concurrency boundary Stability Related Source",
           "path": "apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx",
           "symbols": "",
-          "summary": "Version: 0.2.26",
-          "body": "- Version: 0.2.26 @agentskit/observability/langfuse is a drop-in observer that ships AgentsKit trace spans to . LLM generations become generation nodes; tool calls and planning steps become span nodes. Token usage and error state are captured on each node end. Cost is not normalized by this adapter, and HITL is not represented until the canonical event contract includes it. Version: 0.2.20 The langfuse npm package is an optional peer — loaded dynamically at runtime. If it is absent or no API keys are provided, the observer is a no-op and does not throw. When to reach for it - You already use Langfuse for prompt management, user feedback, or trace analytics. - You want per-session or per-user trace grouping without manual instrumentation. - You want token and latency metadata in the Langfuse UI alongside your AgentsKit runs. Best fit - Add alongside consoleLogger or opentelemetry — observers compose and do not conflict. - Pair with to turn Langfuse trace failures into repeatable eval cases. - Pair with to trace multi-step agent workflows end-to-end. Install Hello world Keys are read from LANGFUSE PUBLIC KEY and LANGFUSE SECRET KEY env vars. No config object required for the common case. Public API langfuse(config?): Observer Config field Type Default Notes --- --- --- --- publicKey string? LANGFUSE PUBLIC KEY env Langfuse project public key secretKey string? LANGFUSE SECRET KEY env Langfuse project secret key baseUrl string? LANGFUSE HOST env, then https://cloud.langfuse.com Self-hosted base URL release string? LANGFUSE RELEASE env Deployment release tag environment string? LANGFUSE ENVIRONMENT env e.g. production , staging sessionId string? — Groups traces into a Langfuse session userId string? — Associates traces with a Langfuse user tags string[]? — Arbitrary trace tags flushAt number? 15 Batch size before auto-flush flushInterval number? 1000 Auto-flush interval in ms onError (error: unknown) = void \\ Promise<void — Isolated sink for SDK/configuration failures; callback failures are swallowed Returns an Observer that can be passed directly to observers in createRuntime , useChat , or any other AgentsKit entry point that accepts observers. LangfuseConfig Named export — the type of the config object above. Useful when building helper factories. Per-session tracing Self-hosted Langfuse Configuration Env var Purpose --- --- LANGFUSE PUBLIC KEY Langfuse project public key (client-safe) LANGFUSE SECRET KEY Langfuse project secret key (server only) LANGFUSE HOST Self-hosted base URL; defaults to Langfuse cloud LANGFUSE RELEASE Release tag attached to every trace LANGFUSE ENVIRONMENT Environment tag (e.g. production ) Explicit config fields always take precedence over env vars. Concurrency boundary Use one observer instance per serialized run. The current canonical event contract has no run/correlation ID, so shared observers do not claim correctness for concurrent or interleaved runs. Delegated work is represented as flat independent spans until an explicit parent correlation contract is available. Stability - Version: inherited from @agentskit/observability ; this is a private workspace implementation exposed through its /langfuse subpath. - Tier: beta - Contract: Observer interface stable; Langfuse SDK peer-resolved at runtime. Related - — core observer package this adapter plugs into - — convert trace failures into eval cases - - Source npm: · repo:",
+          "summary": "Version: 0.2.27",
+          "body": "- Version: 0.2.27 @agentskit/observability/langfuse is a drop-in observer that ships AgentsKit trace spans to . LLM generations become generation nodes; tool calls and planning steps become span nodes. Token usage and error state are captured on each node end. Cost is not normalized by this adapter, and HITL is not represented until the canonical event contract includes it. Version: 0.2.20 The langfuse npm package is an optional peer — loaded dynamically at runtime. If it is absent or no API keys are provided, the observer is a no-op and does not throw. When to reach for it - You already use Langfuse for prompt management, user feedback, or trace analytics. - You want per-session or per-user trace grouping without manual instrumentation. - You want token and latency metadata in the Langfuse UI alongside your AgentsKit runs. Best fit - Add alongside consoleLogger or opentelemetry — observers compose and do not conflict. - Pair with to turn Langfuse trace failures into repeatable eval cases. - Pair with to trace multi-step agent workflows end-to-end. Install Hello world Keys are read from LANGFUSE PUBLIC KEY and LANGFUSE SECRET KEY env vars. No config object required for the common case. Public API langfuse(config?): Observer Config field Type Default Notes --- --- --- --- publicKey string? LANGFUSE PUBLIC KEY env Langfuse project public key secretKey string? LANGFUSE SECRET KEY env Langfuse project secret key baseUrl string? LANGFUSE HOST env, then https://cloud.langfuse.com Self-hosted base URL release string? LANGFUSE RELEASE env Deployment release tag environment string? LANGFUSE ENVIRONMENT env e.g. production , staging sessionId string? — Groups traces into a Langfuse session userId string? — Associates traces with a Langfuse user tags string[]? — Arbitrary trace tags flushAt number? 15 Batch size before auto-flush flushInterval number? 1000 Auto-flush interval in ms onError (error: unknown) = void \\ Promise<void — Isolated sink for SDK/configuration failures; callback failures are swallowed Returns an Observer that can be passed directly to observers in createRuntime , useChat , or any other AgentsKit entry point that accepts observers. LangfuseConfig Named export — the type of the config object above. Useful when building helper factories. Per-session tracing Self-hosted Langfuse Configuration Env var Purpose --- --- LANGFUSE PUBLIC KEY Langfuse project public key (client-safe) LANGFUSE SECRET KEY Langfuse project secret key (server only) LANGFUSE HOST Self-hosted base URL; defaults to Langfuse cloud LANGFUSE RELEASE Release tag attached to every trace LANGFUSE ENVIRONMENT Environment tag (e.g. production ) Explicit config fields always take precedence over env vars. Concurrency boundary Use one observer instance per serialized run. The current canonical event contract has no run/correlation ID, so shared observers do not claim correctness for concurrent or interleaved runs. Delegated work is represented as flat independent spans until an explicit parent correlation contract is available. Stability - Version: inherited from @agentskit/observability ; this is a private workspace implementation exposed through its /langfuse subpath. - Tier: beta - Contract: Observer interface stable; Langfuse SDK peer-resolved at runtime. Related - — core observer package this adapter plugs into - — convert trace failures into eval cases - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61006,7 +61022,7 @@
             }
           ]
         },
-        "contentHash": "c49ede22b1482583096c04ce6c7ab8862a3b7c6bf2d6914a608c891df418e439",
+        "contentHash": "acb7a721d0c3fea44e57fa429d9778edd8e7312a82b37e5b79599a180416bdb0",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61028,7 +61044,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/observability.mdx",
           "symbols": "",
           "summary": "@agentskit/observability is the package you add when you no longer want to guess what your agent did. It turns opaque runs into inspectable traces, costs, audit events, and live debugging streams.",
-          "body": "@agentskit/observability is the package you add when you no longer want to guess what your agent did. It turns opaque runs into inspectable traces, costs, audit events, and live debugging streams. When to reach for it - You need console / LangSmith / OpenTelemetry logging. - You want an offline HTML trace viewer. - You want a hard dollar ceiling ( costGuard ). - You want a tamper-evident audit log (SOC 2 / HIPAA friendly). - You want a live devtools feed (browser extension compatible). Best fit - Add this before opening an agent to real users or production workloads. - Pair with to trace multi-step behavior. - Pair with when you want to turn failures into repeatable quality work. - Pair with concepts when the workflow has compliance or audit requirements. Install Hello world For many teams, this is the point where an agent stops feeling magical and starts feeling operable. Surface - Loggers: consoleLogger · langsmith · opentelemetry . - Tracing: createTraceTracker · createFileTraceSink · buildTraceReport · renderTraceViewerHtml . - Durable budgets: CostStore · createInMemoryCostStore · recordCostSpend · postgresCostStore ( /postgres ) · costStoreContract ( /cost-store-contract ). - Cost + tokens: costGuard · priceFor · computeCost · DEFAULT PRICES · approximateCounter · countTokens · countTokensDetailed · createProviderCounter . - Audit: createSignedAuditLog · createInMemoryAuditStore . - Devtools: createDevtoolsServer · toSseFrame . Recipes - - - - Stability - Version: 0.12.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/observability is the package you add when you no longer want to guess what your agent did. It turns opaque runs into inspectable traces, costs, audit events, and live debugging streams. When to reach for it - You need console / LangSmith / OpenTelemetry logging. - You want an offline HTML trace viewer. - You want a hard dollar ceiling ( costGuard ). - You want a tamper-evident audit log (SOC 2 / HIPAA friendly). - You want a live devtools feed (browser extension compatible). Best fit - Add this before opening an agent to real users or production workloads. - Pair with to trace multi-step behavior. - Pair with when you want to turn failures into repeatable quality work. - Pair with concepts when the workflow has compliance or audit requirements. Install Hello world For many teams, this is the point where an agent stops feeling magical and starts feeling operable. Surface - Loggers: consoleLogger · langsmith · opentelemetry . - Tracing: createTraceTracker · createFileTraceSink · buildTraceReport · renderTraceViewerHtml . - Durable budgets: CostStore · createInMemoryCostStore · recordCostSpend · postgresCostStore ( /postgres ) · costStoreContract ( /cost-store-contract ). - Cost + tokens: costGuard · priceFor · computeCost · DEFAULT PRICES · approximateCounter · countTokens · countTokensDetailed · createProviderCounter . - Audit: createSignedAuditLog · createInMemoryAuditStore . - Devtools: createDevtoolsServer · toSseFrame . Recipes - - - - Stability - Version: 0.13.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61125,7 +61141,7 @@
             }
           ]
         },
-        "contentHash": "ef1b809976de1c818cb8a99839f929d4a9fa5bf13b634598684abffad563e6aa",
+        "contentHash": "4daf7c1c8db1aafdf83d90c55dee29c94d5172669877f29c66dc241df1aafa0f",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61326,7 +61342,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/rag.mdx",
           "symbols": "",
           "summary": "You want retrieval-augmented generation in a few lines.",
-          "body": "When to reach for it - You want retrieval-augmented generation in a few lines. - You need rerankers (BM25, Voyage, Jina, or a custom function) or hybrid vector+keyword search. - You want document loaders for URL / GitHub / Notion / Confluence / Google Drive / PDF / S3 / GCS / Dropbox / OneDrive. Install Hello world Surface - createRAG({ embed, store, chunkSize, chunkOverlap, topK, threshold }) . - chunkText — standalone splitter. - @agentskit/rag/markdown — parseNote · splitFrontmatter · extractWikilinks · stripWikilinks · sections · findSection · parseTable · normalize · plain for Markdown notes (frontmatter, wikilinks, heading sections, GitHub tables). - createRerankedRetriever · createHybridRetriever · bm25Score · bm25Rerank . - Loaders: loadUrl · loadGitHubFile · loadGitHubTree · loadNotionPage · loadConfluencePage · loadGoogleDriveFile · loadPdf · loadS3 · loadGcs · loadDropbox · loadOneDrive . Remote loaders use a finite timeout and response-size limit by default. Configure timeoutMs , maxResponseBytes , and signal when needed. loadUrl requires an HTTPS origin in allowedOrigins and rejects redirects. Recipes - - - - Stability - Version: 0.6.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - Source npm: · repo: Partition-bound retrieval createPartitionedRAG composes createRAG without changing the core Retriever contract or existing createRAG behavior. Bind each instance to a non-empty partition at construction: Ingest overrides the reserved akPartition metadata field and namespaces chunk IDs with a JSON tuple, so identical document IDs in different partitions do not overwrite or delete each other's chunks. Document metadata, including documentId , remains available. Returned chunk IDs are opaque namespaced IDs. Every search forwards a core VectorFilter equality predicate and verifies partition metadata locally, rejecting foreign or missing metadata even when a store ignores filters. The wrapper makes one store search, requests at most topK overfetchFactor candidates, examines at most that many, and returns at most topK matches. The factor is an integer from 1 to 16 (default 4); unsafe candidate limits throw. Per-call search limits and thresholds still work. The required onDiagnostics callback runs for every successful store search, including empty results, with partition , requestedTopK , candidateLimit , examined , rejected , returned , and recall: 'potentially-reduced' . This is conservative: bounded over-fetch can miss matching chunks, and a store can internally filter a bounded global candidate pool without revealing exclusions (as fileVectorMemory does). Short results never justify returning another partition's chunks. Callback exceptions reject the search. Stores must faithfully preserve chunk metadata and IDs. Missing metadata fails closed, but fabricated or incorrectly reassigned partition metadata cannot be detected. This is retrieval isolation, not an authorization boundary against untrusted writers or stores. Use physically separate stores or a backend with trusted partition enforcement when that boundary or exhaustive recall is needed. Backend-internal over-fetch is outside the wrapper's candidate bound.",
+          "body": "When to reach for it - You want retrieval-augmented generation in a few lines. - You need rerankers (BM25, Voyage, Jina, or a custom function) or hybrid vector+keyword search. - You want document loaders for URL / GitHub / Notion / Confluence / Google Drive / PDF / S3 / GCS / Dropbox / OneDrive. Install Hello world Surface - createRAG({ embed, store, chunkSize, chunkOverlap, topK, threshold }) . - chunkText — standalone splitter. - @agentskit/rag/markdown — parseNote · splitFrontmatter · extractWikilinks · stripWikilinks · sections · findSection · parseTable · normalize · plain for Markdown notes (frontmatter, wikilinks, heading sections, GitHub tables). - createRerankedRetriever · createHybridRetriever · bm25Score · bm25Rerank . - Loaders: loadUrl · loadGitHubFile · loadGitHubTree · loadNotionPage · loadConfluencePage · loadGoogleDriveFile · loadPdf · loadS3 · loadGcs · loadDropbox · loadOneDrive . Remote loaders use a finite timeout and response-size limit by default. Configure timeoutMs , maxResponseBytes , and signal when needed. loadUrl requires an HTTPS origin in allowedOrigins and rejects redirects. Recipes - - - - Stability - Version: 0.7.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - Source npm: · repo: Partition-bound retrieval createPartitionedRAG composes createRAG without changing the core Retriever contract or existing createRAG behavior. Bind each instance to a non-empty partition at construction: Ingest overrides the reserved akPartition metadata field and namespaces chunk IDs with a JSON tuple, so identical document IDs in different partitions do not overwrite or delete each other's chunks. Document metadata, including documentId , remains available. Returned chunk IDs are opaque namespaced IDs. Every search forwards a core VectorFilter equality predicate and verifies partition metadata locally, rejecting foreign or missing metadata even when a store ignores filters. The wrapper makes one store search, requests at most topK overfetchFactor candidates, examines at most that many, and returns at most topK matches. The factor is an integer from 1 to 16 (default 4); unsafe candidate limits throw. Per-call search limits and thresholds still work. The required onDiagnostics callback runs for every successful store search, including empty results, with partition , requestedTopK , candidateLimit , examined , rejected , returned , and recall: 'potentially-reduced' . This is conservative: bounded over-fetch can miss matching chunks, and a store can internally filter a bounded global candidate pool without revealing exclusions (as fileVectorMemory does). Short results never justify returning another partition's chunks. Callback exceptions reject the search. Stores must faithfully preserve chunk metadata and IDs. Missing metadata fails closed, but fabricated or incorrectly reassigned partition metadata cannot be detected. This is retrieval isolation, not an authorization boundary against untrusted writers or stores. Use physically separate stores or a backend with trusted partition enforcement when that boundary or exhaustive recall is needed. Backend-internal over-fetch is outside the wrapper's candidate bound.",
           "aliases": "document human"
         },
         "graph": {
@@ -61388,7 +61404,7 @@
             }
           ]
         },
-        "contentHash": "c76ee7a3326f5d0fc84ed374abf4e6d79b4649dd329eb731c722a00e9455df5c",
+        "contentHash": "ef350a62dd4e53a8a700209289ece498cad7ec12adec467e9c47da3d86611a7f",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61410,7 +61426,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/react-native.mdx",
           "symbols": "",
           "summary": "Provider credentials must stay behind a server/edge endpoint; Expo and other mobile environment variables are public and must not contain long-lived API keys.",
-          "body": "Install Hello world Provider credentials must stay behind a server/edge endpoint; Expo and other mobile environment variables are public and must not contain long-lived API keys. Surface - useChat(config): ChatReturn — shares the core state/action contract; component props use React Native primitives and do not promise full web parity. Siblings · · · · · Stability - Version: 0.5.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
+          "body": "Install Hello world Provider credentials must stay behind a server/edge endpoint; Expo and other mobile environment variables are public and must not contain long-lived API keys. Surface - useChat(config): ChatReturn — shares the core state/action contract; component props use React Native primitives and do not promise full web parity. Siblings · · · · · Stability - Version: 0.5.10 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61432,7 +61448,7 @@
             }
           ]
         },
-        "contentHash": "63c0330c2ac7f272a142a6892869095135793c29a24032d64321b85109a4f6b4",
+        "contentHash": "d87cb4cb70523fe8788862d0292bf5fd464b546bae84a186558d2b3799ef86c0",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61454,7 +61470,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/react.mdx",
           "symbols": "",
           "summary": "@agentskit/react is the fastest path from the AgentsKit contracts to a browser product. It gives you the same core chat model as the rest of the ecosystem, but with React ergonomics and headless UI primitives.",
-          "body": "@agentskit/react is the fastest path from the AgentsKit contracts to a browser product. It gives you the same core chat model as the rest of the ecosystem, but with React ergonomics and headless UI primitives. When to reach for it - You're building a web chat UI in React. - You want streaming + tool calls + memory + skills in one hook. - You want headless components with data-ak- attributes so you style everything with your own CSS / Tailwind / shadcn. Best fit - Start here when you need a customer-facing or internal chat experience in React. - Pair with for provider flexibility. - Pair with and when the chat needs to do real work. - Pair with and when the assistant needs longer context. Install Hello world This is the UI entry point for the ecosystem, not a separate product. The same contracts keep working as you add runtime, tools, memory, and production layers. Surface - useChat(config): ChatReturn . - <ChatContainer , <Message , <InputBar , <ToolCallView , <ToolConfirmation , <ThinkingIndicator , <CodeBlock , <Markdown . - Theme: @agentskit/react/theme . Siblings (shared state/action contract) · · · · · Recipes - - - - Stability - Version: 0.8.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/react is the fastest path from the AgentsKit contracts to a browser product. It gives you the same core chat model as the rest of the ecosystem, but with React ergonomics and headless UI primitives. When to reach for it - You're building a web chat UI in React. - You want streaming + tool calls + memory + skills in one hook. - You want headless components with data-ak- attributes so you style everything with your own CSS / Tailwind / shadcn. Best fit - Start here when you need a customer-facing or internal chat experience in React. - Pair with for provider flexibility. - Pair with and when the chat needs to do real work. - Pair with and when the assistant needs longer context. Install Hello world This is the UI entry point for the ecosystem, not a separate product. The same contracts keep working as you add runtime, tools, memory, and production layers. Surface - useChat(config): ChatReturn . - <ChatContainer , <Message , <InputBar , <ToolCallView , <ToolConfirmation , <ThinkingIndicator , <CodeBlock , <Markdown . - Theme: @agentskit/react/theme . Siblings (shared state/action contract) · · · · · Recipes - - - - Stability - Version: 0.8.10 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61501,7 +61517,7 @@
             }
           ]
         },
-        "contentHash": "21f699615d3ee6117b7677f145804fbcb2a4d026d2882f5417648415850ec615",
+        "contentHash": "a4bb6bf390e371c106562e75f09cdd7cf038954689170a7a843db6e275874be2",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61677,7 +61693,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/runtime.mdx",
           "symbols": "",
           "summary": "@agentskit/runtime is where AgentsKit stops being “a chat integration” and becomes an agent system. This package owns the multi-step loop that lets a model inspect, act, reflect, and continue.",
-          "body": "@agentskit/runtime is where AgentsKit stops being “a chat integration” and becomes an agent system. This package owns the multi-step loop that lets a model inspect, act, reflect, and continue. When to reach for it - You want an agent without a UI. - You need durable execution (resume after a crash). - You want ready-made multi-agent topologies. - You want to race adapters ( speculate ) or run on cron / webhooks. Best fit - Start here when the task is more than one model call. - Pair with when the model needs to do work in the world. - Pair with when context needs to persist across runs. - Pair with before trusting the system in production. Install Hello world That one run() call is the center of the agent stack. Everything else in the ecosystem layers into or around it. Surface - createRuntime(config) — headless agent with tools / memory / skills / observers. - createSharedContext — typed context across tools. - createDurableRunner + createInMemoryStepLog / createFileStepLog — Temporal-style step log. - supervisor · swarm · hierarchical · blackboard — multi-agent topologies. - speculate — race adapters, abort losers. - createCronScheduler + createWebhookHandler — background agents. Recipes - - - - - Stability - Version: 0.11.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/runtime is where AgentsKit stops being “a chat integration” and becomes an agent system. This package owns the multi-step loop that lets a model inspect, act, reflect, and continue. When to reach for it - You want an agent without a UI. - You need durable execution (resume after a crash). - You want ready-made multi-agent topologies. - You want to race adapters ( speculate ) or run on cron / webhooks. Best fit - Start here when the task is more than one model call. - Pair with when the model needs to do work in the world. - Pair with when context needs to persist across runs. - Pair with before trusting the system in production. Install Hello world That one run() call is the center of the agent stack. Everything else in the ecosystem layers into or around it. Surface - createRuntime(config) — headless agent with tools / memory / skills / observers. - createSharedContext — typed context across tools. - createDurableRunner + createInMemoryStepLog / createFileStepLog — Temporal-style step log. - supervisor · swarm · hierarchical · blackboard — multi-agent topologies. - speculate — race adapters, abort losers. - createCronScheduler + createWebhookHandler — background agents. Recipes - - - - - Stability - Version: 0.11.2 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61744,7 +61760,7 @@
             }
           ]
         },
-        "contentHash": "acb56d36d6ab24f5fd464105ff92e0823db0ea6d39a94e7b3ce670cce8adb9c5",
+        "contentHash": "e036a7f1a0e0d59200f2d478a727021591159b71b63373ae9fbd2072dfb5dbef",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61766,7 +61782,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/sandbox.mdx",
           "symbols": "",
           "summary": "@agentskit/sandbox is the package for the moment when your agent needs to execute code or touch risky tools and “please be careful” is no longer enough. It adds a policy layer between the model and dangerous capabilities.",
-          "body": "@agentskit/sandbox is the package for the moment when your agent needs to execute code or touch risky tools and “please be careful” is no longer enough. It adds a policy layer between the model and dangerous capabilities. When to reach for it - You want the model to run code safely. - You want a policy layer over every tool (allow / deny / require-sandbox / validators). Best fit - Add this before enabling shell, code execution, or broad filesystem access. - Pair with to wrap risky tool surfaces. - Pair with when autonomous agents can take meaningful actions. - Pair with production security guidance before public rollout. Install Hello world This is one of the clearest boundaries between a capable agent demo and a production-safe agent system. Surface - createSandbox(config?) — default backend probes E2B. - sandboxTool() — ready-made code execution tool (js / python). - createE2BBackend(config) — BYO E2B. - createMandatorySandbox({ sandbox, policy }) . Recipes - Stability - Version: 0.6.12 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - — tools to wrap. Source npm: · repo:",
+          "body": "@agentskit/sandbox is the package for the moment when your agent needs to execute code or touch risky tools and “please be careful” is no longer enough. It adds a policy layer between the model and dangerous capabilities. When to reach for it - You want the model to run code safely. - You want a policy layer over every tool (allow / deny / require-sandbox / validators). Best fit - Add this before enabling shell, code execution, or broad filesystem access. - Pair with to wrap risky tool surfaces. - Pair with when autonomous agents can take meaningful actions. - Pair with production security guidance before public rollout. Install Hello world This is one of the clearest boundaries between a capable agent demo and a production-safe agent system. Surface - createSandbox(config?) — default backend probes E2B. - sandboxTool() — ready-made code execution tool (js / python). - createE2BBackend(config) — BYO E2B. - createMandatorySandbox({ sandbox, policy }) . Recipes - Stability - Version: 0.6.13 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - — tools to wrap. Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61803,7 +61819,7 @@
             }
           ]
         },
-        "contentHash": "71a06422fcac73ee88ae3d23750bb1009d6025a0397f8fbd1ebcfbe1381d0944",
+        "contentHash": "12e59255242a51d64b5606cf98a4337af41a3b6d23b4ffa97e284f1084fc25a5",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61825,7 +61841,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/skills.mdx",
           "symbols": "",
           "summary": "You want one of 26 contract-tested personas across general, engineering, data, support, or regulated-domain workflows.",
-          "body": "When to reach for it - You want one of 26 contract-tested personas across general, engineering, data, support, or regulated-domain workflows. - You want to publish + install skills by semver. Install Hello world Surface - Ready-made: 26 named SkillDefinition exports; use getBuiltinSkills() for the canonical defensive catalog. - Composition: composeSkills · getBuiltinSkills · listSkills . - Marketplace: createSkillRegistry · parseSemver · compareSemver · matchesRange . Tool and delegate fields are declarative references. Wire the corresponding registries in the consuming runtime; importing a skill alone does not install or execute those capabilities. Recipes - - - Stability - Version: 0.9.9 - Tier: beta - Contract: SkillDefinition semantics pinned by ADR 0005; package conveniences remain beta. - Roadmap: stable surface proposed in ; ADR 0024 evidence remains mandatory. Related - - - Source npm: · repo:",
+          "body": "When to reach for it - You want one of 26 contract-tested personas across general, engineering, data, support, or regulated-domain workflows. - You want to publish + install skills by semver. Install Hello world Surface - Ready-made: 26 named SkillDefinition exports; use getBuiltinSkills() for the canonical defensive catalog. - Composition: composeSkills · getBuiltinSkills · listSkills . - Marketplace: createSkillRegistry · parseSemver · compareSemver · matchesRange . Tool and delegate fields are declarative references. Wire the corresponding registries in the consuming runtime; importing a skill alone does not install or execute those capabilities. Recipes - - - Stability - Version: 0.9.10 - Tier: beta - Contract: SkillDefinition semantics pinned by ADR 0005; package conveniences remain beta. - Roadmap: stable surface proposed in ; ADR 0024 evidence remains mandatory. Related - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61872,7 +61888,7 @@
             }
           ]
         },
-        "contentHash": "b9aaa3a1ba16d15fc211e7bd644727228d57701427bf0e0d1cba49a6aec0d318",
+        "contentHash": "e31dbc239e29ad8ae34513b06423ce5f047265ae5eaef29511fa1fc407dfa3cc",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61894,7 +61910,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/solid.mdx",
           "symbols": "",
           "summary": "useChat(config): ChatReturn — Solid hook backed by createStore + onCleanup.",
-          "body": "Install Hello world Surface - useChat(config): ChatReturn — Solid hook backed by createStore + onCleanup . - SSR module boundary: import the hook from @agentskit/solid/use-chat ; the root entry also includes browser UI components. The shared state/action contract is portable; Solid component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.5.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
+          "body": "Install Hello world Surface - useChat(config): ChatReturn — Solid hook backed by createStore + onCleanup . - SSR module boundary: import the hook from @agentskit/solid/use-chat ; the root entry also includes browser UI components. The shared state/action contract is portable; Solid component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.5.10 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61916,7 +61932,7 @@
             }
           ]
         },
-        "contentHash": "98b1da134eb7bbc88694b281a148cca5746ffe86a2f3259cfad296c7cdb82f15",
+        "contentHash": "2e10d7d9889e291f39aec531fd3a8dee2da718097cd6c49d4b5c30439deb409b",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61938,7 +61954,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/statechart.mdx",
           "symbols": "",
           "summary": "@agentskit/statechart is a small, framework-neutral primitive for interactive agent experiences that need explicit states and resumable JSON snapshots.",
-          "body": "@agentskit/statechart is a small, framework-neutral primitive for interactive agent experiences that need explicit states and resumable JSON snapshots. When to reach for it - A chat or agent UI has deterministic interaction states. - The same state contract must work across web, native, server, and terminal hosts. - A host needs to save and restore interaction state with its own storage. - Replay must not depend on hidden clocks or random IDs. Use instead when you need durable execution, tools, effects, or flow orchestration. Install Core operations - defineStatechart — validate and freeze a trusted definition. - createStatechartInstance — create validated JSON state with a host-supplied ID and timestamp. - transitionStatechart — run one pure transition and receive an accepted / rejected result. - serializeStatechart — produce a versioned JSON-compatible snapshot. - restoreStatechart — validate an unknown snapshot against the trusted definition and injected context parser. - notifyStatechartObserver — deliver a completed result without putting observation inside the transition core. Ownership boundary The package owns transition and snapshot semantics. Hosts own persistence, event delivery, deduplication, clocks, IDs, and side effects. Framework adapters own rendering. Runtime owns agent and tool execution. Stability - Version: 0.3.1 - Current release: 0.3.0 - Tier: beta - Contracts: , Related - - - Source repo:",
+          "body": "@agentskit/statechart is a small, framework-neutral primitive for interactive agent experiences that need explicit states and resumable JSON snapshots. When to reach for it - A chat or agent UI has deterministic interaction states. - The same state contract must work across web, native, server, and terminal hosts. - A host needs to save and restore interaction state with its own storage. - Replay must not depend on hidden clocks or random IDs. Use instead when you need durable execution, tools, effects, or flow orchestration. Install Core operations - defineStatechart — validate and freeze a trusted definition. - createStatechartInstance — create validated JSON state with a host-supplied ID and timestamp. - transitionStatechart — run one pure transition and receive an accepted / rejected result. - serializeStatechart — produce a versioned JSON-compatible snapshot. - restoreStatechart — validate an unknown snapshot against the trusted definition and injected context parser. - notifyStatechartObserver — deliver a completed result without putting observation inside the transition core. Ownership boundary The package owns transition and snapshot semantics. Hosts own persistence, event delivery, deduplication, clocks, IDs, and side effects. Framework adapters own rendering. Runtime owns agent and tool execution. Stability - Version: 0.3.2 - Current release: 0.3.0 - Tier: beta - Contracts: , Related - - - Source repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -62000,7 +62016,7 @@
             }
           ]
         },
-        "contentHash": "9a18c08d7f9891e5d25c5d8831a7f6f3f6cd9a65e2463b381f8bee5166932541",
+        "contentHash": "7c6b9fdf86438681027bdb62ed5b1bb88be3d06a2d2e6db203eb62360ddef1e0",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -62022,7 +62038,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/svelte.mdx",
           "symbols": "",
           "summary": "createChatStore(config): SvelteChatStore — Readable<ChatState> + action methods + destroy(). send accepts string | ContentPart[]; decide(id, 'approve' | 'deny', reason?) forwards durable decisions to core with a configured decisionStore. Both actions reject after destroy().",
-          "body": "Install Hello world Surface - createChatStore(config): SvelteChatStore — Readable<ChatState + action methods + destroy() . send accepts string ContentPart[] ; decide(id, 'approve' 'deny', reason?) forwards durable decisions to core with a configured decisionStore . Both actions reject after destroy() . The shared state/action contract is portable; Svelte component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.5.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
+          "body": "Install Hello world Surface - createChatStore(config): SvelteChatStore — Readable<ChatState + action methods + destroy() . send accepts string ContentPart[] ; decide(id, 'approve' 'deny', reason?) forwards durable decisions to core with a configured decisionStore . Both actions reject after destroy() . The shared state/action contract is portable; Svelte component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.6.0 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -62044,7 +62060,7 @@
             }
           ]
         },
-        "contentHash": "372bfce01f26f0589d9417bc067fd53b277d2e2c854655b09039bd672d77486b",
+        "contentHash": "7bb3adde2a6b7058f26494178710cd4d0f30d075e942da1e8363238b20ccf05b",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -62066,7 +62082,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/templates.mdx",
           "symbols": "",
           "summary": "import { ContributeCallout } from '@/components/contribute/contribute-callout'",
-          "body": "import { ContributeCallout } from '@/components/contribute/contribute-callout' Authoring toolkit for generating AgentsKit extensions: validated ToolDefinition / SkillDefinition / AdapterFactory objects and on-disk scaffolds (package.json, tsup, tests, README). Depends only on . For full application starters, see . @agentskit/templates is the lower-level programmatic toolkit for extension packages — it is not the implementation behind agentskit init . When to use - You publish custom tools , skills , adapters , memory , embedders , or flows as standalone packages. - You want consistent blueprints (tsup, vitest, TypeScript) across internal plugins. - You need runtime validation before registering a template with a runtime or marketplace. Install Public API Export Role -------- ------ createToolTemplate(config) Build a ToolDefinition with validation createSkillTemplate(config) Build a SkillDefinition with validation + metadata passthrough createAdapterTemplate(config) Build an AdapterFactory + name + optional capabilities scaffold(config) Write a full package directory (async, atomic) validateScaffoldConfig(config) Validate scaffold input before write validateToolTemplate / validateSkillTemplate / validateAdapterTemplate Assert well-formed definitions SCAFFOLD TYPES Tuple of the eight allowed types createToolTemplate ToolTemplateConfig extends a partial tool with required name and optional description , schema (JSON Schema object), execute , tags , category , requiresConfirmation , init , dispose , and base merge. Validation throws if name / description are empty after trim, schema is not a plain object, or execute is not a function. createSkillTemplate Requires name , description , and systemPrompt (trim non-empty). Optional: examples , tools , delegates , finite temperature , metadata , onActivate , base . createAdapterTemplate Requires non-empty name and createSource . Optional capabilities passthrough. scaffold Creates a directory join(dir, name) with: - package.json , tsconfig.json , tsup.config.ts - src/index.ts (named-export stub — no default export in source) - tests/index.test.ts (contract tests) - README.md (+ flow.yaml for type: 'flow' ) ScaffoldType : 'tool' 'skill' 'adapter' 'memory-vector' 'memory-chat' 'flow' 'embedder' 'browser-adapter' . Security: config validated first; unscoped kebab-case names only; existing destinations fail unless overwrite: true ; symlink destination roots rejected; sibling staging + atomic rename; cleanup on failure. Returned paths are final, not staging. Dependencies generated: @agentskit/core ^1.0.0 always; flow also @agentskit/runtime ^0.10.0 . No wildcards. No unused adapters/memory deps. Scoped packages are not supported in this beta; a future minor may add them with an explicit migration. Register scaffolded packages like any other tool, skill, or adapter via or . Troubleshooting Error Cause ------- -------- Tool name must be a non-empty string Pass a trim-non-empty name . requires a schema JSON Schema plain object required (not null/array). Skill ... systemPrompt Skills must define behavior via systemPrompt . Adapter ... createSource Factory must expose createSource(request) . Destination already exists Choose another name/dir or pass overwrite: true . Refusing to ... symlink Point dir at a real directory, not a symlink target name. not a safe unscoped npm package id Use kebab-case like my-search (no scopes yet). See also · · ( @agentskit/templates ) · · · · <ContributeCallout pkg=\"templates\" / Stability - Version: 0.5.9 - Tier: beta (hardened scaffold/validation surface; not yet stable) - Promotion path: - Contract: evolving under beta semver (breaking changes allowed in minors with CHANGELOG notes)",
+          "body": "import { ContributeCallout } from '@/components/contribute/contribute-callout' Authoring toolkit for generating AgentsKit extensions: validated ToolDefinition / SkillDefinition / AdapterFactory objects and on-disk scaffolds (package.json, tsup, tests, README). Depends only on . For full application starters, see . @agentskit/templates is the lower-level programmatic toolkit for extension packages — it is not the implementation behind agentskit init . When to use - You publish custom tools , skills , adapters , memory , embedders , or flows as standalone packages. - You want consistent blueprints (tsup, vitest, TypeScript) across internal plugins. - You need runtime validation before registering a template with a runtime or marketplace. Install Public API Export Role -------- ------ createToolTemplate(config) Build a ToolDefinition with validation createSkillTemplate(config) Build a SkillDefinition with validation + metadata passthrough createAdapterTemplate(config) Build an AdapterFactory + name + optional capabilities scaffold(config) Write a full package directory (async, atomic) validateScaffoldConfig(config) Validate scaffold input before write validateToolTemplate / validateSkillTemplate / validateAdapterTemplate Assert well-formed definitions SCAFFOLD TYPES Tuple of the eight allowed types createToolTemplate ToolTemplateConfig extends a partial tool with required name and optional description , schema (JSON Schema object), execute , tags , category , requiresConfirmation , init , dispose , and base merge. Validation throws if name / description are empty after trim, schema is not a plain object, or execute is not a function. createSkillTemplate Requires name , description , and systemPrompt (trim non-empty). Optional: examples , tools , delegates , finite temperature , metadata , onActivate , base . createAdapterTemplate Requires non-empty name and createSource . Optional capabilities passthrough. scaffold Creates a directory join(dir, name) with: - package.json , tsconfig.json , tsup.config.ts - src/index.ts (named-export stub — no default export in source) - tests/index.test.ts (contract tests) - README.md (+ flow.yaml for type: 'flow' ) ScaffoldType : 'tool' 'skill' 'adapter' 'memory-vector' 'memory-chat' 'flow' 'embedder' 'browser-adapter' . Security: config validated first; unscoped kebab-case names only; existing destinations fail unless overwrite: true ; symlink destination roots rejected; sibling staging + atomic rename; cleanup on failure. Returned paths are final, not staging. Dependencies generated: @agentskit/core ^1.0.0 always; flow also @agentskit/runtime ^0.10.0 . No wildcards. No unused adapters/memory deps. Scoped packages are not supported in this beta; a future minor may add them with an explicit migration. Register scaffolded packages like any other tool, skill, or adapter via or . Troubleshooting Error Cause ------- -------- Tool name must be a non-empty string Pass a trim-non-empty name . requires a schema JSON Schema plain object required (not null/array). Skill ... systemPrompt Skills must define behavior via systemPrompt . Adapter ... createSource Factory must expose createSource(request) . Destination already exists Choose another name/dir or pass overwrite: true . Refusing to ... symlink Point dir at a real directory, not a symlink target name. not a safe unscoped npm package id Use kebab-case like my-search (no scopes yet). See also · · ( @agentskit/templates ) · · · · <ContributeCallout pkg=\"templates\" / Stability - Version: 0.5.10 - Tier: beta (hardened scaffold/validation surface; not yet stable) - Promotion path: - Contract: evolving under beta semver (breaking changes allowed in minors with CHANGELOG notes)",
           "aliases": "document human"
         },
         "graph": {
@@ -62148,7 +62164,7 @@
             }
           ]
         },
-        "contentHash": "ab7be64918967558608f217cd6d69b6f461ca84adc094f69d164b98c5cc155be",
+        "contentHash": "10706eeb53c2aa2f2ff85ef946531adc81acb2aad00b92750f27ca72c373ca6e",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -62170,7 +62186,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/tools.mdx",
           "symbols": "",
           "summary": "@agentskit/tools is what turns a model from “something that answers” into “something that can actually do work”. It is the action layer of the ecosystem.",
-          "body": "@agentskit/tools is what turns a model from “something that answers” into “something that can actually do work”. It is the action layer of the ecosystem. When to reach for it - You need ready-made tools (web search, file I/O, shell). - You want GitHub / Linear / Slack / Notion / Stripe / Postgres / S3 / … - You want to speak MCP (consume or publish). Best fit - Start here when prompts alone stop being enough. - Pair with for autonomous agents and workflows. - Pair with when you want tool-aware chat surfaces and confirmations. - Pair with and security recipes before enabling risky actions. Install Hello world Most production agents become valuable only once this layer is in place. Surface - Main: webSearch · fetchUrl · filesystem · shell · defineZodTool . - /integrations : github · linear · slack · notion · discord · gmail · googleCalendar · stripe · postgres · s3 · firecrawl · reader · documentParsers · openaiImages · elevenlabs · whisper · deepgram · maps · weather · coingecko · browserAgent . - /mcp : createMcpClient · createMcpServer · toolsFromMcpClient · createStdioTransport · createInMemoryTransportPair . safeFetch(input, init?, policy?) defaults to denying non-public addresses, uses Node DNS to check every hostname answer, and re-gates each manual redirect. Vetted internal targets require an explicit allowedHosts entry or allowPrivateHosts: true ; the DNS preflight does not prevent rebinding. The isPrivateIPv4 and isPrivateHost helpers retain decimal handling for zero-padded dotted quads; safeFetch applies WHATWG URL canonicalization first. webSearch() reads provider responses through @agentskit/net with a 2 MiB default byte cap ( maxResponseBytes ) and cancels the stream on overflow. Recipes - - - - - - Stability - Version: 0.13.14 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
+          "body": "@agentskit/tools is what turns a model from “something that answers” into “something that can actually do work”. It is the action layer of the ecosystem. When to reach for it - You need ready-made tools (web search, file I/O, shell). - You want GitHub / Linear / Slack / Notion / Stripe / Postgres / S3 / … - You want to speak MCP (consume or publish). Best fit - Start here when prompts alone stop being enough. - Pair with for autonomous agents and workflows. - Pair with when you want tool-aware chat surfaces and confirmations. - Pair with and security recipes before enabling risky actions. Install Hello world Most production agents become valuable only once this layer is in place. Surface - Main: webSearch · fetchUrl · filesystem · shell · defineZodTool . - /integrations : github · linear · slack · notion · discord · gmail · googleCalendar · stripe · postgres · s3 · firecrawl · reader · documentParsers · openaiImages · elevenlabs · whisper · deepgram · maps · weather · coingecko · browserAgent . - /mcp : createMcpClient · createMcpServer · toolsFromMcpClient · createStdioTransport · createInMemoryTransportPair . safeFetch(input, init?, policy?) defaults to denying non-public addresses, uses Node DNS to check every hostname answer, and re-gates each manual redirect. Vetted internal targets require an explicit allowedHosts entry or allowPrivateHosts: true ; the DNS preflight does not prevent rebinding. The isPrivateIPv4 and isPrivateHost helpers retain decimal handling for zero-padded dotted quads; safeFetch applies WHATWG URL canonicalization first. webSearch() reads provider responses through @agentskit/net with a 2 MiB default byte cap ( maxResponseBytes ) and cancels the stream on overflow. Recipes - - - - - - Stability - Version: 0.13.15 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -62367,7 +62383,7 @@
             }
           ]
         },
-        "contentHash": "1926701dc379e3eaee8b94876ffff67e2a2fac797adef09f1e35265d6e8b05b2",
+        "contentHash": "f191d6d99e58f14104dcb644021c464391703a59bdc726c7f47f02681d60ada2",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -62389,7 +62405,7 @@
           "path": "apps/docs-next/content/docs/reference/packages/vue.mdx",
           "symbols": "",
           "summary": "useChat(config): ChatReturn — reactive via Vue's reactive() + auto-cleanup on scope dispose.",
-          "body": "Install Hello world Surface - useChat(config): ChatReturn — reactive via Vue's reactive() + auto-cleanup on scope dispose. - <ChatContainer :config=\"...\" — headless container using data-ak- attributes. The shared state/action contract is portable; Vue component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.5.9 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
+          "body": "Install Hello world Surface - useChat(config): ChatReturn — reactive via Vue's reactive() + auto-cleanup on scope dispose. - <ChatContainer :config=\"...\" — headless container using data-ak- attributes. The shared state/action contract is portable; Vue component props and exports are binding-specific. This package does not promise full React component parity. Siblings · · · · · Stability - Version: 0.5.10 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -62411,7 +62427,7 @@
             }
           ]
         },
-        "contentHash": "1d37433f2377c39748a6ee6a5d68755f70974f23aedefc6488a1b20159973ec4",
+        "contentHash": "57a94acb35de8805ad81d8ba9e1eb18b62c529495c0e8edd7e378aba31f7de2e",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -63037,7 +63053,7 @@
             }
           ]
         },
-        "contentHash": "7fbecf065621c8074f25b15d6589aa68628998d04950e879b6ef71709a2dc1e0",
+        "contentHash": "6c0deafff5744176d351b6c42aae19daab54983b5e6f1ee286ec8d0547ca21b5",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -64046,7 +64062,7 @@
             }
           ]
         },
-        "contentHash": "a81ab301cdde13f04a8b4ae4d85328ef1e54c65083b5f12624a8e58dc5c54648",
+        "contentHash": "78128080568b626ffc2fa51aecb47bd4117340555df3724668407d9727ad5123",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -72468,7 +72484,7 @@
           "path": "docs/architecture/adrs/0044-durable-cost-store.md",
           "symbols": "",
           "summary": "Status: Proposed (implementation authorized; evidence review remains external)",
-          "body": "ADR 0044 — Durable tenant cost store - Status : Proposed (implementation authorized; evidence review remains external) - Date : 2026-10-09 Context The cost guards in @agentskit/observability keep spend in closure state. On hosts that run many short-lived instances, each instance starts at zero, so a tenant budget is never reached and concurrent requests cannot see each other. A budget that must hold needs spend recorded outside the process, and admission that is atomic under concurrency. Decision Add a CostStore port to @agentskit/observability with four tenant-scoped operations: reserve , commit , release and window . - reserve holds an estimate against a cap passed by the caller. The cap is an argument, not stored state, so the host reads it from its own plan data. - commit replaces the hold with the real spend and appends usage ledger rows. - release returns the hold when the call fails. - Each operation is idempotent per (tenant, reservationId) . Settling an unknown, released or committed reservation raises CostReservationError . - Spend is grouped by an accounting windowKey , the UTC month by default. Ship two implementations: createInMemoryCostStore for development and tests, and postgresCostStore on the optional /postgres subpath, with BYO Drizzle and node-postgres, exported schema and idempotent DDL. Each PostgreSQL operation is one statement; admission is the conditional update spent + reserved + amount <= cap , so no explicit transaction is required and the store works with a pooled or a per-request connection. Export costStoreContract on /cost-store-contract as the framework-free suite any implementation must pass. costGuard , multiTenantCostGuard and createAdvancedCostGuard accept an optional store . With it they write each priced llm:end to the store and trip on the stored window total. Without it they keep the existing process-local behavior and warn once when NODE ENV=production . Consequences - Existing guard options and behavior are unchanged when store is omitted, apart from the production warning. - With a store, a guard budget applies to the store's current accounting window instead of the lifetime of one guard instance. - Guards observe spend after it happens and write to the store asynchronously, so they bound overshoot to in-flight calls. Hosts that must refuse a request before it spends call reserve directly. - commit records the real spend even above the reserved estimate. The cap is a hard ceiling on reservations, not on provider usage already incurred. - drizzle-orm and pg become optional peers, required only by /postgres . - The advanced guard's rolling window caps ( perMinute , perDay , custom) remain process-local; only the overall budget reads the store. Alternatives considered - Store the cap in the cost tables. Rejected: plan limits belong to the host's billing data, and a second copy would drift. - Interactive transaction with SELECT … FOR UPDATE . Rejected: it needs a dedicated connection per reservation and does not suit pooled HTTP drivers. - Record spend only, without reservations. Rejected: concurrent requests would all pass a check made before any of them commits.",
+          "body": "ADR 0044 — Durable tenant cost store - Status : Proposed (implementation authorized; evidence review remains external) - Date : 2026-10-09 Context The cost guards in @agentskit/observability keep spend in closure state. On hosts that run many short-lived instances, each instance starts at zero, so a tenant budget is never reached and concurrent requests cannot see each other. A budget that must hold needs spend recorded outside the process, and admission that is atomic under concurrency. Decision Add a CostStore port to @agentskit/observability with four tenant-scoped operations: reserve , commit , release and window . - reserve holds an estimate against a cap passed by the caller. The cap is an argument, not stored state, so the host reads it from its own plan data. - commit replaces the hold with the real spend and appends usage ledger rows. - release returns the hold when the call fails. - Each operation is idempotent per (tenant, reservationId) . Settling an unknown, released or committed reservation raises CostReservationError . - Spend is grouped by an accounting windowKey , the UTC month by default. Ship two implementations: createInMemoryCostStore for development and tests, and postgresCostStore on the optional /postgres subpath, with BYO Drizzle and node-postgres, exported schema and idempotent DDL. Each PostgreSQL operation is one statement; admission is the conditional update spent + reserved + amount <= cap , so no explicit transaction is required and the store works with a pooled or a per-request connection. Export costStoreContract on /cost-store-contract as the framework-free suite any implementation must pass. costGuard , multiTenantCostGuard and createAdvancedCostGuard accept an optional store . With it they write each priced llm:end to the store and trip on the stored window total. Without it they keep the existing process-local behavior and warn once in a production runtime: NODE ENV=production , or Cloudflare Workers when NODE ENV is not set. warnWithoutStore overrides the detection. A store write that fails goes to onError ; a guard without onError logs the first failure once. Reservations have no time to live in the store. expirePostgresCostReservations releases the ones still held and older than a cutoff the host chooses; the host schedules it. recordCostSpend releases its own hold when the commit fails. Consequences - Existing guard options and behavior are unchanged when store is omitted, apart from the production warning. - With a store, a guard budget applies to the store's current accounting window instead of the lifetime of one guard instance. - Guards observe spend after it happens and write to the store asynchronously, so they bound overshoot to in-flight calls. Hosts that must refuse a request before it spends call reserve directly. - commit records the real spend even above the reserved estimate. The cap is a hard ceiling on reservations, not on provider usage already incurred. - drizzle-orm and pg become optional peers, required only by /postgres . - A released reservation id is not admitted again: reserve raises AK COST RESERVATION RELEASED . Replaying it as admitted would let the call run without a hold and then fail to commit. - A reservation abandoned by a crashed process stays held until the host runs the expiry. A cutoff shorter than a call still in flight releases its hold, and that call's commit is then rejected. - reserve must run outside a host transaction or inside a savepoint. A concurrent duplicate id ends in a unique violation that the store absorbs, which still aborts an enclosing transaction. - The CommonJS build of each subpath has its own copy of CostReservationError . The contract suite and hosts match errors by code . - The advanced guard's rolling window caps ( perMinute , perDay , custom) remain process-local; only the overall budget reads the store. Alternatives considered - Store the cap in the cost tables. Rejected: plan limits belong to the host's billing data, and a ",
           "aliases": "document human"
         },
         "graph": {
@@ -72516,7 +72532,7 @@
             }
           ]
         },
-        "contentHash": "9f0c1e6420206c1f6a813589cd231af59b72d22fb68a895a6bafed1d6f1b8203",
+        "contentHash": "6bbcb59353c45545b97728da92df345307d0642d32582c46c731d181393edf64",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73028,7 +73044,7 @@
           "path": "docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/carousel-storyboard.md",
           "symbols": "",
           "summary": "Format: 5-frame carousel / 15s silent GIF",
-          "body": "Storyboard — Connect AgentsKit to your coding agent Format: 5-frame carousel / 15s silent GIF 1. Hook — \"Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\" 2. Install — show codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search 3. Copy fixture — show the agent.ts path 4. Run — terminal output includes the expected ready line 5. Next step — point to Registry / Playbook / Chat without inventing counts Accessibility: every frame needs alt text; prefer reduced-motion static frames. No auto-publish — storyboard is a draft for human design review.",
+          "body": "Storyboard — Connect AgentsKit to your coding agent Format: 5-frame carousel / 15s silent GIF 1. Hook — \"Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\" 2. Install — show codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search 3. Copy fixture — show the agent.ts path 4. Run — terminal output includes the expected ready line 5. Next step — point to Registry / Playbook / Chat without inventing counts Accessibility: every frame needs alt text; prefer reduced-motion static frames. No auto-publish — storyboard is a draft for human design review.",
           "aliases": "document human"
         },
         "graph": {
@@ -73039,7 +73055,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "2c3613d9b17a0ab1ae8076e55a0f885968b06b566cc34d446b7d9094b58988e7",
+        "contentHash": "98a26068fa0636dd6d79762b1b3606616e5b823f49ee4f86718f147446fdf8fd",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73078,7 +73094,7 @@
             }
           ]
         },
-        "contentHash": "e13286472557d766283be88550b5db9fc0ee583c4521ab5a236b7848e5399dfb",
+        "contentHash": "90e92591b6bc6b4b991729c7eda9deb21e37bbe5ff482f40c713691e82438c4d",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73111,7 +73127,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "2a89209aae9627e8fc934630b0fb27e7861283fbb558f9bfd97c661128ef725b",
+        "contentHash": "0c786965cab639ef1728feb7b4355039740f699811aaac3bd529a33e9ea89fc6",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73199,7 +73215,7 @@
           "path": "docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/social-thread.md",
           "symbols": "",
           "summary": "1/ Connect AgentsKit to your coding agent",
-          "body": "1/ Connect AgentsKit to your coding agent Why it exists: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host. 2/ The path: 1. codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search 2. claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search 3. pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts 4. pnpm --filter @agentskit/mcp smoke:published 3/ Claims (generated, not hand-typed): - 22 packages - 50 integrations 4/ Next products: - AgentsKit: Build agents without gluing many libraries together. - AgentsKit Registry: Copy ready-made agents and own the source. - AgentsKit Chat: Define one agent experience and deliver it across interfaces. 5/ This thread is a draft atom. Publishing requires explicit human approval.",
+          "body": "1/ Connect AgentsKit to your coding agent Why it exists: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host. 2/ The path: 1. codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search 2. claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search 3. pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts 4. pnpm --filter @agentskit/mcp smoke:published 3/ Claims (generated, not hand-typed): - 22 packages - 50 integrations 4/ Next products: - AgentsKit: Build agents without gluing many libraries together. - AgentsKit Registry: Copy ready-made agents and own the source. - AgentsKit Chat: Define one agent experience and deliver it across interfaces. 5/ This thread is a draft atom. Publishing requires explicit human approval.",
           "aliases": "document human"
         },
         "graph": {
@@ -73210,7 +73226,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "8807e7a4ace7ed7bb5f507aa82b588068ebcbb97cc12aba5331aa56100c32f6f",
+        "contentHash": "05e7837ff75b4b53c65053d8843d6d575ba35bcc12d4ee157c5d5d04cd6b9310",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -79755,7 +79771,7 @@
             }
           ]
         },
-        "contentHash": "d140f0cf8621abad4dab454537ebab797c96004bec9da916c579828703254ff5",
+        "contentHash": "4306e774390c1c87a40c6cb508eab760bdb0886c2d0b5ac5ce0180e90ba9c574",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -79849,7 +79865,7 @@
             }
           ]
         },
-        "contentHash": "44223f4f7f15d48dc74085432f785211877a9a1a84ab09f2946ae9fac44aacb5",
+        "contentHash": "3097341299c6313b1a1b6cfafddb2149faf3137477da3d953f6a9495dc8cd8f7",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -80692,7 +80708,7 @@
             }
           ]
         },
-        "contentHash": "f784f385ea347b6b418c6d84d67f7c6a100a21806c880640f9edbfdcf1635221",
+        "contentHash": "87237cd2b57ba5c3b22587d59167010673075706fdfde7ff9b2eb344783f4de1",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -150566,7 +150582,7 @@
             }
           ]
         },
-        "contentHash": "e937255484065fc80d6a9d5508b896caf651356180b91ddd8142c59d6b8db3bb",
+        "contentHash": "999745c52fe403d21ecd122d746b8f48f0c519ed557c4f9a58d19a52341d729f",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -150673,7 +150689,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "13a202839963acc9607784c70d35d9d2cc1d2fcde7c32e170708d97729201fc6",
+        "contentHash": "7d404d08513159d609e6039a3ed49c4a0a0294d9d629515d51d3dbdcadb989d7",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -151165,7 +151181,7 @@
             }
           ]
         },
-        "contentHash": "76f2f46976359b2d942e774336963692bffb0c5865fb52845db70bfa91e3b063",
+        "contentHash": "090e00a1a759cff1439b2606e7571e19d27cd6f92f9d20284d712caeb8f7f363",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159149,7 +159165,7 @@
             }
           ]
         },
-        "contentHash": "75f302e345d2a4bd69fbdf08b4693eedaa50a59c2c6a019fcc1774913d5ed12f",
+        "contentHash": "362f6a41a929494e30cb3dcba902a6d67c15ef163fe9e83d7524706528d6308c",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159293,7 +159309,7 @@
             }
           ]
         },
-        "contentHash": "6577cedb8e48993548baefaae96fbc910d157fe3e5d355345ff3d286c0fa5dc3",
+        "contentHash": "ebf4e6e4938bc3f0ea7a9791cf6a6168b97d45380428630d1885007b90ddb989",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159364,7 +159380,7 @@
             }
           ]
         },
-        "contentHash": "9a078a80a71859f4200812350e14376f90fd49f0b602bed5e006e4054c18259b",
+        "contentHash": "f790b633d6eaff1bc10da4fd92388afcf8dd51ae9cd2222c434409a11cba9aad",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159554,7 +159570,7 @@
             }
           ]
         },
-        "contentHash": "dbed63fe3d0288ce8719ea849709b60870774e988be501b23fe69a731785355d",
+        "contentHash": "7a574710105c8a6d41a5f3615e20ca713dd8db041620b1fcefa2648ead6e089d",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159628,7 +159644,7 @@
             }
           ]
         },
-        "contentHash": "1e5f787b952cbbd9bdfd9b809cbdf7a8c315ee9a3def164285420482c228f189",
+        "contentHash": "f523e7509cb4e0f9c8901d9ad0dcf071142527b03e4d139b693e4b647e0598b5",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -159657,7 +159673,9 @@
           "monthlyWindowKey",
           "recordCostSpend",
           "validateCostCommit",
+          "validateCostRelease",
           "validateCostReserve",
+          "validateCostWindow",
           "warnProcessLocalCostState"
         ],
         "tags": [
@@ -159669,7 +159687,7 @@
           "title": "cost-store.ts",
           "headings": "",
           "path": "packages/observability/src/cost-store.ts",
-          "symbols": "CostCommitInput CostReleaseInput CostReservationError CostReserveInput CostReserveResult CostSpendRecord CostStore CostStoreOptions CostUsageEntry CostWindow CostWindowInput InMemoryCostStore InMemoryLedgerRow createInMemoryCostStore describeCostWindow monthlyWindowKey recordCostSpend validateCostCommit validateCostReserve warnProcessLocalCostState",
+          "symbols": "CostCommitInput CostReleaseInput CostReservationError CostReserveInput CostReserveResult CostSpendRecord CostStore CostStoreOptions CostUsageEntry CostWindow CostWindowInput InMemoryCostStore InMemoryLedgerRow createInMemoryCostStore describeCostWindow monthlyWindowKey recordCostSpend validateCostCommit validateCostRelease validateCostReserve validateCostWindow warnProcessLocalCostState",
           "summary": "",
           "body": "",
           "aliases": "module ts src"
@@ -159744,7 +159762,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "000d000f32f8ff879d6372400a7620b9574392e10947972ffa252c1339d51dfe",
+        "contentHash": "a4de0beaaa030a3cccabedb0b68e616d9333997bde4ca6ce6f2251203b2bae81",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -160573,7 +160591,9 @@
         "title": "postgres.ts",
         "aliases": [],
         "symbols": [
+          "ExpirePostgresCostReservationsOptions",
           "PostgresCostStoreOptions",
+          "expirePostgresCostReservations",
           "postgresCostMigrationSql",
           "postgresCostReservationTable",
           "postgresCostStore",
@@ -160589,7 +160609,7 @@
           "title": "postgres.ts",
           "headings": "",
           "path": "packages/observability/src/postgres.ts",
-          "symbols": "PostgresCostStoreOptions postgresCostMigrationSql postgresCostReservationTable postgresCostStore postgresCostWindowTable postgresUsageLedgerTable",
+          "symbols": "ExpirePostgresCostReservationsOptions PostgresCostStoreOptions expirePostgresCostReservations postgresCostMigrationSql postgresCostReservationTable postgresCostStore postgresCostWindowTable postgresUsageLedgerTable",
           "summary": "",
           "body": "",
           "aliases": "module ts src"
@@ -160634,7 +160654,7 @@
             }
           ]
         },
-        "contentHash": "c717f504c6644d0f7d1db9017415638bc8f1889e8e974545c508aa31484da167",
+        "contentHash": "08672acc3b8820e6a96bdd2e30521771091138ce0a5484ab1749e5d931bbbe23",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -161369,6 +161389,39 @@
         "provenance": "observed",
         "confidence": "observed"
       },
+      {
+        "id": "module:packages/observability/tests/cjs-smoke.cjs",
+        "kind": "module",
+        "path": "packages/observability/tests/cjs-smoke.cjs",
+        "title": "cjs-smoke.cjs",
+        "aliases": [],
+        "tags": [
+          "module",
+          "cjs",
+          "tests"
+        ],
+        "fields": {
+          "title": "cjs-smoke.cjs",
+          "headings": "",
+          "path": "packages/observability/tests/cjs-smoke.cjs",
+          "symbols": "",
+          "summary": "",
+          "body": "",
+          "aliases": "module cjs tests"
+        },
+        "graph": {
+          "pagerank": 0,
+          "inboundLinks": 0,
+          "coveredBy": [],
+          "mentionedBy": [],
+          "packageId": "package:@agentskit/observability",
+          "inbound": [],
+          "outbound": []
+        },
+        "contentHash": "56ca4f72a7dbbd0a9eafecc08c5c3678b1e767896657fee8d5525b1c4e814427",
+        "provenance": "observed",
+        "confidence": "observed"
+      },
       {
         "id": "module:packages/observability/tests/console-logger.test.ts",
         "kind": "module",
@@ -161710,7 +161763,7 @@
             }
           ]
         },
-        "contentHash": "3dbca16e7462eb3d4946f12a3730c0396d4a0762cc414572d1e6d820e0d74288",
+        "contentHash": "f0bd5884a15e5771ce295772ddca61703b9d9f4b7edf47188486a18d7d254e98",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -161795,7 +161848,7 @@
             }
           ]
         },
-        "contentHash": "447ac4782c545122cb71dfc17c03751951c9e62ec542b47d91d2df3e2768e3f8",
+        "contentHash": "3a284757bbc12d9d061ea11791f9b4e8e9cb9b9a69e4cf78b630f88aaa4f9008",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -161840,7 +161893,7 @@
             }
           ]
         },
-        "contentHash": "263a1943013345075c1080f4b31228f7fe934bc31feb05d2f0b5c0ab9fd42f1d",
+        "contentHash": "3a7ffc7e3fca98ac3da8ed711ee91f5c57e4d24b8d744f4e87c24d286ba6c9a6",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -162092,7 +162145,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "f3d3e7a3eae7a1c1d5cddb2bf506ea81fdc0fff40056edbd57505a02e1b4611b",
+        "contentHash": "5950721a855c0536a6d54b89b013d507fb567b63df59a2cbbd669a04b3ce5289",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -162134,7 +162187,7 @@
             }
           ]
         },
-        "contentHash": "407e64dc7114efd7920d36a27a747899ccb99afbf79db8a86c2ec404df756f12",
+        "contentHash": "9e78a9f4cc0757bbe13b5f4e8f149d62c4f8d6df9e6e82678e1f5d3c34a1428b",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -162167,7 +162220,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "7a7a6a85faffe15c65b44898f95505d629d8aa57a2d7c3750c0a187a15f92bce",
+        "contentHash": "dbd4f1b1f4c518fdc519e586f9b2655a6856d036d084504108b47fc410b72a69",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -162207,7 +162260,7 @@
             }
           ]
         },
-        "contentHash": "8fb55eaf341dfedbc79f7ce382f60e785b3954afb810e36048803a02c7c1feb7",
+        "contentHash": "301ce5fb51c3d0910a5eba60c067ab36ee65c26c879b2882cef099bb6fed0156",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -162247,7 +162300,7 @@
             }
           ]
         },
-        "contentHash": "9f3255e0a36e2632e7119af79ca31ee09386584d3b1f1d6e9df665710ac3f76a",
+        "contentHash": "21268cb747a51e617a2679d73362734f99c6f338e9766d7e72f97f4d805aba0f",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -193558,7 +193611,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "007f4f057d1350928e92c2191139a32e7e62a13bb57a5c63f68ea59041fe0a30",
+        "contentHash": "b1164d3b3208a1b262fd83e32b625b86347fde0fcea35f6e20f15386ad23b728",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "adapters"
@@ -193821,7 +193874,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "3ee52469206c0f914f51021c50a8f5bb7c8d188dcf982689d8730bacf54685ff",
+        "contentHash": "dd4b10bcb4e2207eb11f1e95d940b00634c97f1fae6be8161a7f180398b0c062",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "angular"
@@ -194030,7 +194083,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "182ff6e5d5c36122b4a255b861a35194295778289b9e81afbf31f93975ab0b69",
+        "contentHash": "3102994e0e481847532c1cd05781a4e3934d65602b4d31df2f91f044dd2e891b",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "cli"
@@ -194455,7 +194508,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "bd89aeced546ce2e6bf6ad0921a353ee88292c38f0350038b4dec68181232caf",
+        "contentHash": "703be9b2f7d902689dbfc7cc130e1ccb55944eb90da3d18e29a709c21774239d",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "core"
@@ -194562,7 +194615,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "1c6e5be6baf70c4b54eb0c9a1c4d40c85ee94dab70da505ad497a77e5474c24a",
+        "contentHash": "d21f87a35c1c2759357e1710ff15a5a288f7efbf040e7f6e16dd946f011b34a6",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "cross-platform"
@@ -194873,7 +194926,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "d807e0572438666b83aedd614fcc3625a9c41c6c2707fc68ad204114cf0804c2",
+        "contentHash": "326ab16a5cd43e1f3c6e8eff41c207f202a9fc8d8c487e05dd3cf6ff88a0eca8",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "eval"
@@ -194955,7 +195008,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "f3da89db6e109addd0a98d89e41f35629ff9c8e697b22216c6f3d46cb7cafc4e",
+        "contentHash": "2148c472c9a40b3bf764d4317e63682d15500d6ec09210efd891ac645c5d78df",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "eval-braintrust"
@@ -195332,7 +195385,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "36384905390fb131962f1a6f0ea47d9c74c77371bb2820a6ad953500eb4e9704",
+        "contentHash": "6e1bcedffc39b702fd3b941ac3b02f5a5084b4e16617653a76623498e062987a",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "ink"
@@ -195487,7 +195540,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "e833b1373c9faa222ec79b15417ad7e7d9a58de85cd429814a5b472199277ebb",
+        "contentHash": "4f26ce8c48f4d653e4e1f7a1be20e03253830db1815fb5f8519e07f4fcbd566a",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "integrations"
@@ -195513,7 +195566,7 @@
           "path": "packages/mcp",
           "symbols": "",
           "summary": "Serve AgentsKit tools to an MCP host that supports the documented `2024-11-05` tools bridge. A thin bridge over `@agentskit/tools/mcp` (`createMcpServer` + `createStdioTransport`).",
-          "body": "Purpose Serve AgentsKit tools to an MCP host that supports the documented 2024-11-05 tools bridge. A thin bridge over @agentskit/tools/mcp ( createMcpServer + createStdioTransport ). Install Primary exports - createAgentsKitMcpServer({ tools, serverInfo?, onEvent?, transport?, authorizeToolCall? }) — expose McpToolDefinition[] as an MCP server. Defaults to stdio over the current process; pass transport to override (tests / custom hosts). - processStdio() — adapt the Node process into the StdioLikeProcess the transport expects (server receives on stdin, sends on stdout). - createAgentTool({ id, description, systemPrompt, adapter, maxSteps? }) — wrap a whole agent as one MCP tool. The host calls it with a task ; the agent runs server-side on the given adapter and returns the result (\"agents as MCP tools\"). - createTypedAgentTool(config) — typed variant of createAgentTool for an agent input/output contract. - fetchAgent(id, fetchImpl?) — fetch a registry agent definition without executing its source. - fetchAgentSkill(id, fetchImpl?) — fetch a registry agent's runnable skill (hosted index, raw-GitHub fallback); null for tool-composing agents. An optional third argument controls abort, timeout, and response-byte limits. Pass authorizeToolCall to createAgentsKitMcpServer to obtain host approval for tools marked requiresConfirmation . It has the same type as McpServerOptions['authorizeToolCall'] : (tool, args) = boolean Promise<boolean . Arguments are validated first. Only confirmation-required tools invoke the hook; absent hooks, denied approvals, throws, and rejected promises prevent execution. The host must obtain the required human confirmation before returning true ; annotations do not replace that decision. McpToolDefinition (from @agentskit/tools/mcp ) extends the core tool shape with optional title and annotations : title , readOnlyHint , destructiveHint , idempotentHint , and openWorldHint . Server listing and client import preserve these fields, including explicit false hints. Imported titles use the existing maxDescriptionBytes UTF-8 limit; malformed metadata is rejected. Hints remain untrusted advisory data and never grant confirmation or alter execution policy. These fields follow the . They are optional descriptor extensions on the existing 2024-11-05 bridge; older clients may ignore unknown fields. This does not add negotiation or other features from newer protocol revisions. Bin agentskit-mcp — npx -y @agentskit/mcp@0.4.10 --tools fetch,search . Expose whole agents with --agents <ids --provider <p (any of ~20 providers + catalog OpenAI-compatible). Flags: --tools , --fs-root , --sqlite , --allow-shell , --model , --max-steps , --api-key , --base-url . Unknown, duplicate, or valueless flags fail closed. stdout is the JSON-RPC channel; diagnostics and help go to stderr. Pinned coding-agent configs and their evidence levels live in the . Host config fixtures cover Codex, Claude Code/Desktop, Cursor, Cline, and Continue. The local-build and published-package smoke scripts both initialize the shared STDIO path and require the tool list to equal fetch url,web search . Beta boundaries - Tool names are unique and follow the current MCP 1–128 character interoperability grammar. - Server configuration and agent-tool configuration fail with typed AgentsKit diagnostics. - Published tool definitions are snapshotted; observer failures cannot alter protocol behavior. - Agent tasks, step counts, prompts, descriptions, registry IDs, request time, and response bytes are bounded. - Registry source is treated as data and never executed. - The built-in surface is the MCP tools subset over stdio or an injected transport. HTTP/WebSocket, resources, prompts, sampling, auth, rate limiting, and persistence are host responsibilities. Supported protocol matrix Protocol revision Lifecycle Transports Methods Not included --- --- --- --- --- 2024-11-05 Exact-version initialize ; tools/list ; tools/call ; close settles pending calls stdio, in-memor",
+          "body": "Purpose Serve AgentsKit tools to an MCP host that supports the documented 2024-11-05 tools bridge. A thin bridge over @agentskit/tools/mcp ( createMcpServer + createStdioTransport ). Install Primary exports - createAgentsKitMcpServer({ tools, serverInfo?, onEvent?, transport?, authorizeToolCall? }) — expose McpToolDefinition[] as an MCP server. Defaults to stdio over the current process; pass transport to override (tests / custom hosts). - processStdio() — adapt the Node process into the StdioLikeProcess the transport expects (server receives on stdin, sends on stdout). - createAgentTool({ id, description, systemPrompt, adapter, maxSteps? }) — wrap a whole agent as one MCP tool. The host calls it with a task ; the agent runs server-side on the given adapter and returns the result (\"agents as MCP tools\"). - createTypedAgentTool(config) — typed variant of createAgentTool for an agent input/output contract. - fetchAgent(id, fetchImpl?) — fetch a registry agent definition without executing its source. - fetchAgentSkill(id, fetchImpl?) — fetch a registry agent's runnable skill (hosted index, raw-GitHub fallback); null for tool-composing agents. An optional third argument controls abort, timeout, and response-byte limits. Pass authorizeToolCall to createAgentsKitMcpServer to obtain host approval for tools marked requiresConfirmation . It has the same type as McpServerOptions['authorizeToolCall'] : (tool, args) = boolean Promise<boolean . Arguments are validated first. Only confirmation-required tools invoke the hook; absent hooks, denied approvals, throws, and rejected promises prevent execution. The host must obtain the required human confirmation before returning true ; annotations do not replace that decision. McpToolDefinition (from @agentskit/tools/mcp ) extends the core tool shape with optional title and annotations : title , readOnlyHint , destructiveHint , idempotentHint , and openWorldHint . Server listing and client import preserve these fields, including explicit false hints. Imported titles use the existing maxDescriptionBytes UTF-8 limit; malformed metadata is rejected. Hints remain untrusted advisory data and never grant confirmation or alter execution policy. These fields follow the . They are optional descriptor extensions on the existing 2024-11-05 bridge; older clients may ignore unknown fields. This does not add negotiation or other features from newer protocol revisions. Bin agentskit-mcp — npx -y @agentskit/mcp@0.5.0 --tools fetch,search . Expose whole agents with --agents <ids --provider <p (any of ~20 providers + catalog OpenAI-compatible). Flags: --tools , --fs-root , --sqlite , --allow-shell , --model , --max-steps , --api-key , --base-url . Unknown, duplicate, or valueless flags fail closed. stdout is the JSON-RPC channel; diagnostics and help go to stderr. Pinned coding-agent configs and their evidence levels live in the . Host config fixtures cover Codex, Claude Code/Desktop, Cursor, Cline, and Continue. The local-build and published-package smoke scripts both initialize the shared STDIO path and require the tool list to equal fetch url,web search . Beta boundaries - Tool names are unique and follow the current MCP 1–128 character interoperability grammar. - Server configuration and agent-tool configuration fail with typed AgentsKit diagnostics. - Published tool definitions are snapshotted; observer failures cannot alter protocol behavior. - Agent tasks, step counts, prompts, descriptions, registry IDs, request time, and response bytes are bounded. - Registry source is treated as data and never executed. - The built-in surface is the MCP tools subset over stdio or an injected transport. HTTP/WebSocket, resources, prompts, sampling, auth, rate limiting, and persistence are host responsibilities. Supported protocol matrix Protocol revision Lifecycle Transports Methods Not included --- --- --- --- --- 2024-11-05 Exact-version initialize ; tools/list ; tools/call ; close settles pending calls stdio, in-memory",
           "aliases": "mcp @agentskit/mcp package ownership"
         },
         "graph": {
@@ -195648,7 +195701,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "48c9ff77d18d8e907d0b4f99c47f20e53df85ab4ba25499223bc8c86be31fcd9",
+        "contentHash": "dfc303dad9d670aaa921f82ed5baa18f867397a57867826540819e443a5eaece",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "mcp"
@@ -196073,7 +196126,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "9c04e53ca288c601e2b1b6957073505b0beac1c2eda0ceeb0fceeb497435cf09",
+        "contentHash": "60c9a5f7d89613cf23f441f6b2b0b6caa7a197a2761dfed9e733a322f9bca649",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "memory"
@@ -196228,7 +196281,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "f63666e874aa845d83f6418496aaab2a43b5d19419b664df0eda7badd6599bfa",
+        "contentHash": "da8969cda70c8b5ad02952cc9f4cb62876197f5abbea5436e98067a2c9d290d7",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "net"
@@ -196254,7 +196307,7 @@
           "path": "packages/observability",
           "symbols": "",
           "summary": "consoleLogger(config?) — local dev logging.",
-          "body": "Install Primary exports Loggers / tracers - consoleLogger(config?) — local dev logging. - langsmith(config) — LangSmith lifecycle observer ( flush / idempotent shutdown ; optional langsmith peer, lazy). - opentelemetry(config) — OpenTelemetry lifecycle observer (optional OTel peers, lazy). - createTraceTracker({ onSpanStart, onSpanEnd }) — low-level span lifecycle. Cross-repository correlation Every AgentEvent may carry an optional correlation envelope. Set the stable operationId at the host boundary and preserve local runId , sessionId , turnId , actionId , and traceId values when they exist. Observability propagates these values as span attributes without requiring a vendor SDK or a shared runtime. Sinks (batch HTTP exporters) All three return LifecycleObserver ( flush(): Promise<void , idempotent shutdown(): Promise<void ). Shared HttpBatchOptions defaults: batchSize 25, maxQueueSize 1000 (drop-oldest), flushIntervalMs 2000, maxRetries 3, retryBaseDelayMs 100, requestTimeoutMs 10000, plus isolated onError . Single-flight batching, timeout, and retries are best-effort — not an at-least-once guarantee. Await shutdown() on graceful termination. - datadogSink(config) — Datadog Logs intake. - axiomSink(config) — Axiom dataset ingest. - newRelicSink(config) — New Relic Log API. Vendor adapters - @agentskit/observability/langfuse — Langfuse logger backend. Install the optional langfuse peer when using this subpath. Guards + counters - costGuard({ budgetUsd, controller, prices?, unknownModelPolicy?, onCost?, onExceeded?, onError? }) — per-run dollar ceiling; aborts via AbortController when exceeded and fails closed on unknown pricing by default. - multiTenantCostGuard({ budgets, defaultBudgetUsd?, tenantOf?, unknownModelPolicy?, … }) — per-tenant bookkeeping; host enforces (no automatic abort), and unknown pricing marks the tenant blocked by default. - createAdvancedCostGuard({ budgets, caps?, mode?, unknownModelPolicy?, disableRuntime?, alertSinks?, onError?, now? }) — production guard with rolling window caps, 50/80/100 % threshold + forecast alerts, and modes: - warn — observe only. - reject — host must consult isRejected(tenant) (window rejections clear on roll; overall until reset ). Package does not abort the runtime. - kill — requires disableRuntime ; use isDisabled(tenant) ; fail-closed if disable fails. - Accounting is incremental per llm:end / active model ; hostile usage normalizes to zero; zero-budget payloads stay finite. Callback/sink/clock errors are isolated. - consoleAlertSink() , webhookAlertSink({ url, headers, fetch? }) , throttle(sink, windowMs) — built-in alert sinks ( webhook uses injected or globalThis.fetch ; no-op if none). - store?: CostStore on all three guards — durable spend shared by every instance. Without it totals are process-local and the guards warn once when NODE ENV=production . With it, each priced llm:end is written to the store and the budget is checked against the store's current window. - CostStore { reserve, commit, release, window } — tenant-scoped port. reserve({ tenant, reservationId, amountUsd, capUsd?, windowKey? }) returns { ok: true, window } or { ok: false, reason: 'quota exceeded', window } ; commit({ tenant, reservationId, actualUsd, usage? }) writes real spend and ledger rows; release({ tenant, reservationId }) returns the hold. All idempotent per reservationId ; settling an unknown, released or committed reservation throws CostReservationError ( AK COST RESERVATION NOT FOUND / RELEASED / COMMITTED ). - createInMemoryCostStore() — development and test store. recordCostSpend(store, { tenant, costUsd, usage? }) — reserve + commit for spend that already happened. monthlyWindowKey(date) — default window ( YYYY-MM , UTC). - @agentskit/observability/postgres — postgresCostStore({ db }) , postgresCostMigrationSql , and the Drizzle tables postgresCostWindowTable , postgresCostReservationTable , postgresUsageLedgerTable . One statement per operation, cap admitted atomically. Optional pe",
+          "body": "Install Primary exports Loggers / tracers - consoleLogger(config?) — local dev logging. - langsmith(config) — LangSmith lifecycle observer ( flush / idempotent shutdown ; optional langsmith peer, lazy). - opentelemetry(config) — OpenTelemetry lifecycle observer (optional OTel peers, lazy). - createTraceTracker({ onSpanStart, onSpanEnd }) — low-level span lifecycle. Cross-repository correlation Every AgentEvent may carry an optional correlation envelope. Set the stable operationId at the host boundary and preserve local runId , sessionId , turnId , actionId , and traceId values when they exist. Observability propagates these values as span attributes without requiring a vendor SDK or a shared runtime. Sinks (batch HTTP exporters) All three return LifecycleObserver ( flush(): Promise<void , idempotent shutdown(): Promise<void ). Shared HttpBatchOptions defaults: batchSize 25, maxQueueSize 1000 (drop-oldest), flushIntervalMs 2000, maxRetries 3, retryBaseDelayMs 100, requestTimeoutMs 10000, plus isolated onError . Single-flight batching, timeout, and retries are best-effort — not an at-least-once guarantee. Await shutdown() on graceful termination. - datadogSink(config) — Datadog Logs intake. - axiomSink(config) — Axiom dataset ingest. - newRelicSink(config) — New Relic Log API. Vendor adapters - @agentskit/observability/langfuse — Langfuse logger backend. Install the optional langfuse peer when using this subpath. Guards + counters - costGuard({ budgetUsd, controller, prices?, unknownModelPolicy?, onCost?, onExceeded?, onError? }) — per-run dollar ceiling; aborts via AbortController when exceeded and fails closed on unknown pricing by default. - multiTenantCostGuard({ budgets, defaultBudgetUsd?, tenantOf?, unknownModelPolicy?, … }) — per-tenant bookkeeping; host enforces (no automatic abort), and unknown pricing marks the tenant blocked by default. - createAdvancedCostGuard({ budgets, caps?, mode?, unknownModelPolicy?, disableRuntime?, alertSinks?, onError?, now? }) — production guard with rolling window caps, 50/80/100 % threshold + forecast alerts, and modes: - warn — observe only. - reject — host must consult isRejected(tenant) (window rejections clear on roll; overall until reset ). Package does not abort the runtime. - kill — requires disableRuntime ; use isDisabled(tenant) ; fail-closed if disable fails. - Accounting is incremental per llm:end / active model ; hostile usage normalizes to zero; zero-budget payloads stay finite. Callback/sink/clock errors are isolated. - consoleAlertSink() , webhookAlertSink({ url, headers, fetch? }) , throttle(sink, windowMs) — built-in alert sinks ( webhook uses injected or globalThis.fetch ; no-op if none). - store?: CostStore on all three guards — durable spend shared by every instance. Without it totals are process-local and the guards warn once when NODE ENV=production (or on Cloudflare Workers with NODE ENV unset); warnWithoutStore?: boolean overrides that. Store failures go to onError , or to one console.warn when it is absent. With it, each priced llm:end is written to the store and the budget is checked against the store's current window. - CostStore { reserve, commit, release, window } — tenant-scoped port. reserve({ tenant, reservationId, amountUsd, capUsd?, windowKey? }) returns { ok: true, window } or { ok: false, reason: 'quota exceeded', window } ; commit({ tenant, reservationId, actualUsd, usage? }) writes real spend and ledger rows; release({ tenant, reservationId }) returns the hold. All idempotent per reservationId ; settling an unknown, released or committed reservation throws CostReservationError ( AK COST RESERVATION NOT FOUND / RELEASED / COMMITTED ). - createInMemoryCostStore() — development and test store. recordCostSpend(store, { tenant, costUsd, usage? }) — reserve + commit for spend that already happened. monthlyWindowKey(date) — default window ( YYYY-MM , UTC). - @agentskit/observability/postgres — postgresCostStore({ db }) , postgresCostMigrationSql , and the Dr",
           "aliases": "observability @agentskit/observability package ownership"
         },
         "graph": {
@@ -196593,7 +196646,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "f030ef5b129e7deb44709baa55754d0d294f85fcdec3981ad22eafe91ac253d0",
+        "contentHash": "4f22e554c6b109a11ada9de588a00f596ed1a1bccc27469d0525331d44b816e4",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "observability"
@@ -196682,7 +196735,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "8b360b630a3ece5e67d0a85feec394441b262bd1e1117829aea4b3a4c535f1ae",
+        "contentHash": "d232f3cc290e1b805d6735f873b67f55801135f554d30c2a5127025e5a6eec0f",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "observability-langfuse"
@@ -197029,7 +197082,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "123d4ef536efc7296dc97922da67de8852d95f5993e6036139c9dc8ae2b0955a",
+        "contentHash": "c6bc9acad12c0012c4fbb3d7cf0d455b7de0a18a9a4a40b35325d6a185779362",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "rag"
@@ -197454,7 +197507,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "a1d8c04552d06b03af28e42ff585a8717814150c3a1c754a3f91a8ea9655cb27",
+        "contentHash": "397761c892c58f13c2118dd8810a90b527f7f44c3fc778bde5332e6a559ce2fe",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "react"
@@ -197711,7 +197764,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "b37ee239dabf528c2ee0d5b2e0e52a21fa34129d259ad4fce46b5bc5b9bee86d",
+        "contentHash": "098ca49b404753f2d0075d71d5c3b37ddb4e9a1041e3de12b7cdbef1caa9b757",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "react-native"
@@ -198136,7 +198189,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "18d96c319792a6d76adeed5b4dc0c1a1e5c495c52c6f38955d76597e4b13cc50",
+        "contentHash": "f9acf60c076cf7d4127950124d6ce1aad9ecc8b94de1c72079fa58ee7e387e64",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "runtime"
@@ -198417,7 +198470,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "3b081da37207058d1fa9883f3b3637b406daa1916d20460df7ad98ccc6b243b2",
+        "contentHash": "2b8b1c689f86c0329c5bbe5e8ccaf9cc3ff7771ca4e87b4344dfd9c1dda2ac85",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "sandbox"
@@ -198638,7 +198691,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "0ffbc74f7cbd857132dec2e5d75ac7bf3c2c0493e100001769effab7ce7440a9",
+        "contentHash": "0d909b3b4c690866eadb957dc79ca1a8ca6fa981a839d2280f9913bc296cae8b",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "skills"
@@ -198883,7 +198936,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "0c5947149283928e85d49b84e4996fefca57ba472f1d465164138d160bb12be0",
+        "contentHash": "b9bda55e1e279de668d1355b2aadfabd353997b2ebda145adbc2634e730f837d",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "solid"
@@ -199002,7 +199055,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "94ee2e094c5c4da0ff88c214ddda5a1b6a23c5588fda543428d205b2beac7390",
+        "contentHash": "2845efe4453ec2a1be1dc9750556d597a61bea1e47ee4818f0f4565928aaf413",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "statechart"
@@ -199277,7 +199330,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "928e8d0697cec2a10d2e49cf0a8f0987fbf700fc608a4d23fd3840cbf27d377e",
+        "contentHash": "b191cd76150e8e102597a53c4e11c5cc1546a399586b745dcf3c35932aa5c308",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "svelte"
@@ -199450,7 +199503,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "2e9369d5f37ab39d6114eaa230619fa9da87a8e6acdea25fddefdb9173bb5037",
+        "contentHash": "65fafcf0cbb3368966bb81d7801f64b2acfbedaa990b0479bc3e9c94cf0b5c6e",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "templates"
@@ -199875,7 +199928,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "180083715ae5a1c03fb8b4ab3220f6ae761a01eaf32f9debae783f6c2cc361ea",
+        "contentHash": "eaafcb3feda3d95d53cd8c407bd2a7a918f9d20eb48624bf8ccc424597876700",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "tools"
@@ -199994,7 +200047,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "9e450773e54c70a603d0da0509e643d78ae7b79a6d5a4160b52bf067963c923e",
+        "contentHash": "336489409c0d3846d4832dee86a4079175137458eee28baa572290fb55c6703b",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "validation"
@@ -200269,7 +200322,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "92f67398dbceef1dc432e726238cd2c1a53d82acbe6458f98755c367277fe781",
+        "contentHash": "0e79dfd970e8e296f9d269dcd823d44549dc8899cae19701bf62f789458676f3",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "vue"
diff --git a/CITATION.cff b/CITATION.cff
index 76c940f4a..ff68bbbed 100644
--- a/CITATION.cff
+++ b/CITATION.cff
@@ -6,7 +6,7 @@ authors:
   - family-names: "Braun"
     given-names: "Emerson"
     website: "https://github.com/EmersonBraun"
-version: "1.14.1"
+version: "1.15.0"
 license: MIT
 repository-code: "https://github.com/AgentsKit-io/agentskit"
 url: "https://www.agentskit.io"
diff --git a/apps/docs-next/content/docs/for-agents/mcp.mdx b/apps/docs-next/content/docs/for-agents/mcp.mdx
index 1d493896c..5d1cd663c 100644
--- a/apps/docs-next/content/docs/for-agents/mcp.mdx
+++ b/apps/docs-next/content/docs/for-agents/mcp.mdx
@@ -56,7 +56,7 @@ tools bridge. A thin bridge over `@agentskit/tools/mcp` (`createMcpServer` +
 ## Install
 
 ```bash
-npm install @agentskit/mcp@0.4.10
+npm install @agentskit/mcp@0.5.0
 ```
 
 ## Primary exports
@@ -100,7 +100,7 @@ other features from newer protocol revisions.
 
 ## Bin
 
-`agentskit-mcp` — `npx -y @agentskit/mcp@0.4.10 --tools fetch,search`. Expose whole agents
+`agentskit-mcp` — `npx -y @agentskit/mcp@0.5.0 --tools fetch,search`. Expose whole agents
 with `--agents <ids> --provider <p>` (any of ~20 providers + catalog OpenAI-compatible). Flags: `--tools`,
 `--fs-root`, `--sqlite`, `--allow-shell`, `--model`, `--max-steps`, `--api-key`,
 `--base-url`. Unknown, duplicate, or valueless flags fail closed. stdout is the
diff --git a/apps/docs-next/content/docs/reference/packages/adapters.mdx b/apps/docs-next/content/docs/reference/packages/adapters.mdx
index 3bdb2ab75..69fae8b44 100644
--- a/apps/docs-next/content/docs/reference/packages/adapters.mdx
+++ b/apps/docs-next/content/docs/reference/packages/adapters.mdx
@@ -136,7 +136,7 @@ Testing: `mockAdapter` · `recordingAdapter` · `replayAdapter` · `inMemorySink
 
 ## Stability
 
-- **Version:** `0.18.1`
+- **Version:** `0.19.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/angular.mdx b/apps/docs-next/content/docs/reference/packages/angular.mdx
index 86db512f7..8c176a87b 100644
--- a/apps/docs-next/content/docs/reference/packages/angular.mdx
+++ b/apps/docs-next/content/docs/reference/packages/angular.mdx
@@ -57,7 +57,7 @@ export class ChatComponent {
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.5.10`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/cli.mdx b/apps/docs-next/content/docs/reference/packages/cli.mdx
index 069829082..8e6c15071 100644
--- a/apps/docs-next/content/docs/reference/packages/cli.mdx
+++ b/apps/docs-next/content/docs/reference/packages/cli.mdx
@@ -134,7 +134,7 @@ The CLI makes AgentsKit feel like a product, not just a pile of packages. It sho
 
 ## Stability
 
-- **Version:** `0.13.46`
+- **Version:** `0.13.47`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/core.mdx b/apps/docs-next/content/docs/reference/packages/core.mdx
index 79b2d28ce..1644a208b 100644
--- a/apps/docs-next/content/docs/reference/packages/core.mdx
+++ b/apps/docs-next/content/docs/reference/packages/core.mdx
@@ -362,7 +362,7 @@ type SearchArgs = InferSchemaType<typeof schema>
 
 ## Stability
 
-- **Version:** `1.14.1`
+- **Version:** `1.15.0`
 - **Tier:** stable
 - **Contract:** frozen
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/cross-platform.mdx b/apps/docs-next/content/docs/reference/packages/cross-platform.mdx
index 815ae73b1..568bd21fa 100644
--- a/apps/docs-next/content/docs/reference/packages/cross-platform.mdx
+++ b/apps/docs-next/content/docs/reference/packages/cross-platform.mdx
@@ -74,7 +74,7 @@ One native adapter per runtime: `node:child_process` on Node, `Bun.spawn` on Bun
 
 ## Stability
 
-- **Version:** `0.2.1`
+- **Version:** `0.3.0`
 - **Tier:** beta
 - **Contract:** evolving
 
diff --git a/apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx b/apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx
index 9a19dba04..7bee1c32d 100644
--- a/apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx
+++ b/apps/docs-next/content/docs/reference/packages/eval-braintrust.mdx
@@ -24,7 +24,7 @@ docbridge:
       detection: static
 ---
 
-- **Version:** `0.2.27`
+- **Version:** `0.2.28`
 
 `@agentskit/eval/braintrust` connects AgentsKit eval suites to [Braintrust](https://www.braintrust.dev). It runs your test cases through deterministic scorers, optionally ships explicitly selected results to a Braintrust experiment, and surfaces regressions as CI annotations — all without requiring Braintrust to be installed at build time (peer-resolved at runtime). The implementation is a private workspace source package; consumers install the public `@agentskit/eval` package and its `/braintrust` subpath.
 
diff --git a/apps/docs-next/content/docs/reference/packages/eval.mdx b/apps/docs-next/content/docs/reference/packages/eval.mdx
index 281100a48..befbf4539 100644
--- a/apps/docs-next/content/docs/reference/packages/eval.mdx
+++ b/apps/docs-next/content/docs/reference/packages/eval.mdx
@@ -120,7 +120,7 @@ results can be stored directly in metrics; these helpers do not run or judge cas
 
 ## Stability
 
-- **Version:** `0.6.12`
+- **Version:** `0.7.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/ink.mdx b/apps/docs-next/content/docs/reference/packages/ink.mdx
index dc3678c3b..ac8fb2907 100644
--- a/apps/docs-next/content/docs/reference/packages/ink.mdx
+++ b/apps/docs-next/content/docs/reference/packages/ink.mdx
@@ -43,7 +43,7 @@ render(<ChatContainer config={{ adapter: anthropic({ apiKey, model: 'claude-sonn
 
 ## Stability
 
-- **Version:** `0.10.15`
+- **Version:** `0.10.16`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/memory.mdx b/apps/docs-next/content/docs/reference/packages/memory.mdx
index b8524ba6f..eab91045e 100644
--- a/apps/docs-next/content/docs/reference/packages/memory.mdx
+++ b/apps/docs-next/content/docs/reference/packages/memory.mdx
@@ -80,7 +80,7 @@ In most real products, memory is the difference between a clever demo and a usef
 
 ## Stability
 
-- **Version:** `0.11.14`
+- **Version:** `0.12.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/net.mdx b/apps/docs-next/content/docs/reference/packages/net.mdx
index e39ccd3b2..066736abd 100644
--- a/apps/docs-next/content/docs/reference/packages/net.mdx
+++ b/apps/docs-next/content/docs/reference/packages/net.mdx
@@ -82,7 +82,7 @@ await retry(
 
 ## Stability
 
-- **Version:** `0.2.2`
+- **Version:** `0.2.3`
 - **Tier:** beta
 - **Contract:** evolving
 
diff --git a/apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx b/apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx
index c3c0433fc..72bbdc785 100644
--- a/apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx
+++ b/apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx
@@ -23,7 +23,7 @@ docbridge:
       detection: static
 ---
 
-- **Version:** `0.2.26`
+- **Version:** `0.2.27`
 
 `@agentskit/observability/langfuse` is a drop-in observer that ships AgentsKit trace spans to [Langfuse](https://langfuse.com). LLM generations become `generation` nodes; tool calls and planning steps become `span` nodes. Token usage and error state are captured on each node end. Cost is not normalized by this adapter, and HITL is not represented until the canonical event contract includes it.
 
diff --git a/apps/docs-next/content/docs/reference/packages/observability.mdx b/apps/docs-next/content/docs/reference/packages/observability.mdx
index d5c8b2a3d..d88ac26cf 100644
--- a/apps/docs-next/content/docs/reference/packages/observability.mdx
+++ b/apps/docs-next/content/docs/reference/packages/observability.mdx
@@ -82,7 +82,7 @@ For many teams, this is the point where an agent stops feeling magical and start
 
 ## Stability
 
-- **Version:** `0.12.1`
+- **Version:** `0.13.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/rag.mdx b/apps/docs-next/content/docs/reference/packages/rag.mdx
index 22780724a..348c6fc22 100644
--- a/apps/docs-next/content/docs/reference/packages/rag.mdx
+++ b/apps/docs-next/content/docs/reference/packages/rag.mdx
@@ -75,7 +75,7 @@ HTTPS origin in `allowedOrigins` and rejects redirects.
 
 ## Stability
 
-- **Version:** `0.6.1`
+- **Version:** `0.7.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/react-native.mdx b/apps/docs-next/content/docs/reference/packages/react-native.mdx
index 728ba893b..98a3d6a5a 100644
--- a/apps/docs-next/content/docs/reference/packages/react-native.mdx
+++ b/apps/docs-next/content/docs/reference/packages/react-native.mdx
@@ -50,7 +50,7 @@ Provider credentials must stay behind a server/edge endpoint; Expo and other mob
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.5.10`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/react.mdx b/apps/docs-next/content/docs/reference/packages/react.mdx
index eceda2333..1e06ff5ef 100644
--- a/apps/docs-next/content/docs/reference/packages/react.mdx
+++ b/apps/docs-next/content/docs/reference/packages/react.mdx
@@ -72,7 +72,7 @@ This is the UI entry point for the ecosystem, not a separate product. The same c
 
 ## Stability
 
-- **Version:** `0.8.9`
+- **Version:** `0.8.10`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/runtime.mdx b/apps/docs-next/content/docs/reference/packages/runtime.mdx
index e690d2f24..879948a33 100644
--- a/apps/docs-next/content/docs/reference/packages/runtime.mdx
+++ b/apps/docs-next/content/docs/reference/packages/runtime.mdx
@@ -84,7 +84,7 @@ That one `run()` call is the center of the agent stack. Everything else in the e
 
 ## Stability
 
-- **Version:** `0.11.1`
+- **Version:** `0.11.2`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/sandbox.mdx b/apps/docs-next/content/docs/reference/packages/sandbox.mdx
index d660e2e64..6a01312f2 100644
--- a/apps/docs-next/content/docs/reference/packages/sandbox.mdx
+++ b/apps/docs-next/content/docs/reference/packages/sandbox.mdx
@@ -75,7 +75,7 @@ This is one of the clearest boundaries between a capable agent demo and a produc
 
 ## Stability
 
-- **Version:** `0.6.12`
+- **Version:** `0.6.13`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/skills.mdx b/apps/docs-next/content/docs/reference/packages/skills.mdx
index aa40c39b3..7bb207ea6 100644
--- a/apps/docs-next/content/docs/reference/packages/skills.mdx
+++ b/apps/docs-next/content/docs/reference/packages/skills.mdx
@@ -53,7 +53,7 @@ execute those capabilities.
 
 ## Stability
 
-- **Version:** `0.9.9`
+- **Version:** `0.9.10`
 - **Tier:** beta
 - **Contract:** SkillDefinition semantics pinned by ADR 0005; package conveniences remain beta.
 - **Roadmap:** stable surface proposed in [RFC 0009](../../../../../../rfcs/0009-skills-stable.md); ADR 0024 evidence remains mandatory.
diff --git a/apps/docs-next/content/docs/reference/packages/solid.mdx b/apps/docs-next/content/docs/reference/packages/solid.mdx
index 54ae27130..23d93641a 100644
--- a/apps/docs-next/content/docs/reference/packages/solid.mdx
+++ b/apps/docs-next/content/docs/reference/packages/solid.mdx
@@ -42,7 +42,7 @@ component parity.
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.5.10`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/statechart.mdx b/apps/docs-next/content/docs/reference/packages/statechart.mdx
index 8689a3278..fdd8ca034 100644
--- a/apps/docs-next/content/docs/reference/packages/statechart.mdx
+++ b/apps/docs-next/content/docs/reference/packages/statechart.mdx
@@ -38,7 +38,7 @@ The package owns transition and snapshot semantics. Hosts own persistence, event
 
 ## Stability
 
-- **Version:** `0.3.1`
+- **Version:** `0.3.2`
 - **Current release:** `0.3.0`
 - **Tier:** beta
 - **Contracts:** [ADR-0020](../../../../../../docs/architecture/adrs/0020-serializable-interaction-state.md), [ADR-0027](../../../../../../docs/architecture/adrs/0027-statechart-beta-boundaries.md)
diff --git a/apps/docs-next/content/docs/reference/packages/svelte.mdx b/apps/docs-next/content/docs/reference/packages/svelte.mdx
index 9bd6a2c3f..a3dfacbb0 100644
--- a/apps/docs-next/content/docs/reference/packages/svelte.mdx
+++ b/apps/docs-next/content/docs/reference/packages/svelte.mdx
@@ -50,7 +50,7 @@ component parity.
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.6.0`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/templates.mdx b/apps/docs-next/content/docs/reference/packages/templates.mdx
index 5b56b5c62..d6cc738e9 100644
--- a/apps/docs-next/content/docs/reference/packages/templates.mdx
+++ b/apps/docs-next/content/docs/reference/packages/templates.mdx
@@ -173,7 +173,7 @@ Register scaffolded packages like any other tool, skill, or adapter via
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.5.10`
 - **Tier:** beta (hardened scaffold/validation surface; not yet stable)
 - **Promotion path:** [RFC 0014](../../../../../../rfcs/0014-templates-stable.md)
 - **Contract:** evolving under beta semver (breaking changes allowed in minors with CHANGELOG notes)
diff --git a/apps/docs-next/content/docs/reference/packages/tools.mdx b/apps/docs-next/content/docs/reference/packages/tools.mdx
index 5d6e670db..3a574e260 100644
--- a/apps/docs-next/content/docs/reference/packages/tools.mdx
+++ b/apps/docs-next/content/docs/reference/packages/tools.mdx
@@ -103,7 +103,7 @@ default byte cap (`maxResponseBytes`) and cancels the stream on overflow.
 
 ## Stability
 
-- **Version:** `0.13.14`
+- **Version:** `0.13.15`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/packages/vue.mdx b/apps/docs-next/content/docs/reference/packages/vue.mdx
index 152c4c578..012279008 100644
--- a/apps/docs-next/content/docs/reference/packages/vue.mdx
+++ b/apps/docs-next/content/docs/reference/packages/vue.mdx
@@ -43,7 +43,7 @@ are binding-specific. This package does not promise full React component parity.
 
 ## Stability
 
-- **Version:** `0.5.9`
+- **Version:** `0.5.10`
 - **Tier:** beta
 - **Contract:** evolving
 - **Roadmap:** see [packages roadmap](./roadmap) for what this package needs to reach v1.0.
diff --git a/apps/docs-next/content/docs/reference/recipes/coding-agent-mcp.mdx b/apps/docs-next/content/docs/reference/recipes/coding-agent-mcp.mdx
index 0e1fc8350..99e75ae57 100644
--- a/apps/docs-next/content/docs/reference/recipes/coding-agent-mcp.mdx
+++ b/apps/docs-next/content/docs/reference/recipes/coding-agent-mcp.mdx
@@ -35,7 +35,7 @@ host application's UI, login, or trust prompt.
 Add the server with the CLI:
 
 ```bash
-codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 codex mcp list
 ```
 
@@ -44,7 +44,7 @@ Or commit a project-scoped `.codex/config.toml` in a trusted project:
 ```toml
 [mcp_servers.agentskit]
 command = "npx"
-args = ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
 default_tools_approval_mode = "prompt"
 ```
 
@@ -56,7 +56,7 @@ Use `/mcp` to inspect the connected server.
 Create a project-scoped server:
 
 ```bash
-claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 claude mcp list
 ```
 
@@ -68,7 +68,7 @@ The equivalent `.mcp.json` is:
     "agentskit": {
       "type": "stdio",
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
     }
   }
 }
@@ -89,7 +89,7 @@ Create `.cursor/mcp.json`:
     "agentskit": {
       "type": "stdio",
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
     }
   }
 }
@@ -109,7 +109,7 @@ edits `%APPDATA%\Claude\claude_desktop_config.json`.
   "mcpServers": {
     "agentskit": {
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
     }
   }
 }
@@ -128,7 +128,7 @@ Open MCP Servers, choose Configure MCP Servers, and add:
   "mcpServers": {
     "agentskit": {
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"],
+      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"],
       "env": {},
       "disabled": false,
       "autoApprove": []
@@ -151,7 +151,7 @@ Create `.continue/mcpServers/agentskit.yaml`:
 
 ```yaml
 name: AgentsKit MCP
-version: 0.4.10
+version: 0.5.0
 schema: v1
 mcpServers:
   - name: agentskit
@@ -159,7 +159,7 @@ mcpServers:
     command: npx
     args:
       - "-y"
-      - "@agentskit/mcp@0.4.10"
+      - "@agentskit/mcp@0.5.0"
       - "--tools"
       - "fetch,search"
 ```
@@ -174,7 +174,7 @@ server arguments rather than changing host-specific code:
 
 ```bash
 OPENROUTER_API_KEY=your-key \
-  npx -y @agentskit/mcp@0.4.10 \
+  npx -y @agentskit/mcp@0.5.0 \
   --tools fetch,search \
   --agents code-review \
   --provider openrouter \
diff --git a/apps/docs-next/content/docs/reference/recipes/mcp-bridge.mdx b/apps/docs-next/content/docs/reference/recipes/mcp-bridge.mdx
index f801b2fbe..d27b657cb 100644
--- a/apps/docs-next/content/docs/reference/recipes/mcp-bridge.mdx
+++ b/apps/docs-next/content/docs/reference/recipes/mcp-bridge.mdx
@@ -55,7 +55,7 @@ For a ready-made stdio server, install `@agentskit/mcp` instead of writing a
 transport wrapper:
 
 ```bash
-npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 ```
 
 Use the smallest tool set that fits the task. `filesystem`, `sqlite`, and
@@ -71,7 +71,7 @@ Use this JSON in Claude Desktop or `.cursor/mcp.json` in a project:
   "mcpServers": {
     "agentskit": {
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
     }
   }
 }
@@ -84,7 +84,7 @@ Codex uses TOML configuration, not the JSON wrapper above:
 ```toml
 [mcp_servers.agentskit]
 command = "npx"
-args = ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
 ```
 
 ### OpenClaw
@@ -95,7 +95,7 @@ OpenClaw can save and probe the stdio server directly:
 openclaw mcp add agentskit \
   --command npx \
   --arg -y \
-  --arg @agentskit/mcp@0.4.10 \
+  --arg @agentskit/mcp@0.5.0 \
   --arg --tools \
   --arg fetch,search
 openclaw mcp doctor agentskit --probe
diff --git a/apps/docs-next/lib/software-identity.generated.json b/apps/docs-next/lib/software-identity.generated.json
index 983de7c01..82ffead45 100644
--- a/apps/docs-next/lib/software-identity.generated.json
+++ b/apps/docs-next/lib/software-identity.generated.json
@@ -25,7 +25,7 @@
     "url": "https://www.agentskit.io",
     "mainEntityOfPage": "https://www.agentskit.io/docs",
     "license": "https://github.com/AgentsKit-io/agentskit/blob/main/LICENSE",
-    "version": "1.14.1",
+    "version": "1.15.0",
     "dateCreated": "2026-04-02",
     "programmingLanguage": [
       "TypeScript",
@@ -68,7 +68,7 @@
     "description": "A free, MIT-licensed TypeScript ecosystem for production AI agents with replaceable contracts for models, runtimes, tools, memory, RAG, observability, sandboxing, evaluation, and interfaces.",
     "applicationCategory": "DeveloperApplication",
     "operatingSystem": "Cross-platform",
-    "softwareVersion": "1.14.1",
+    "softwareVersion": "1.15.0",
     "offers": {
       "@type": "Offer",
       "price": "0",
diff --git a/codemeta.json b/codemeta.json
index f1f10f00c..30a93bb82 100644
--- a/codemeta.json
+++ b/codemeta.json
@@ -10,7 +10,7 @@
   "url": "https://www.agentskit.io",
   "buildInstructions": "https://www.agentskit.io/docs/get-started/getting-started/installation",
   "license": "https://spdx.org/licenses/MIT",
-  "version": "1.14.1",
+  "version": "1.15.0",
   "dateCreated": "2026-04-02",
   "developmentStatus": "active",
   "programmingLanguage": [
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/atom.json b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/atom.json
index 09036d601..13b9aeae3 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/atom.json
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/atom.json
@@ -43,8 +43,8 @@
       "expectStdoutIncludes": "verified local MCP stdio protocol; cli tools: fetch_url, web_search"
     },
     "commands": [
-      "codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
-      "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
+      "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+      "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
       "pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts",
       "pnpm --filter @agentskit/mcp smoke:published"
     ],
@@ -173,22 +173,22 @@
     }
   ],
   "variants": {
-    "docsPage": "---\ntitle: Connect AgentsKit to your coding agent\ndescription: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n---\n\n# Connect AgentsKit to your coding agent\n\nDevelopers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n## Audience\n\nJavaScript and TypeScript developers using Codex, Claude Code or Desktop, Cursor, Cline, Continue, or another MCP-compatible host\n\n## Commands\n\n```bash\ncodex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search\npnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts\npnpm --filter @agentskit/mcp smoke:published\n```\n\n## Verified claims\n\n- **agentskit/packages**: 22 packages (from ecosystem-claims.json)\n- **agentskit/integrations**: 50 integrations (from ecosystem-claims.json)\n\n## Citations\n\n- [coding-agent MCP host fixture](packages/mcp/fixtures/coding-agent-hosts.ts)\n- [coding-agent MCP recipe](/docs/reference/recipes/coding-agent-mcp)\n- [Codex MCP configuration](https://developers.openai.com/codex/mcp/)\n- [Claude Code MCP configuration](https://docs.anthropic.com/en/docs/claude-code/mcp)\n- [Cursor MCP configuration](https://cursor.com/docs/mcp)\n- [Cline MCP configuration](https://docs.cline.bot/mcp/mcp-overview)\n- [Continue MCP configuration](https://docs.continue.dev/customize/deep-dives/mcp)\n\n## Ecosystem next steps\n\n- [AgentsKit](https://www.agentskit.io/docs) — Build agents without gluing many libraries together.\n- [AgentsKit Registry](https://registry.agentskit.io/docs) — Copy ready-made agents and own the source.\n- [AgentsKit Chat](https://chat.agentskit.io/docs) — Define one agent experience and deliver it across interfaces.\n- [Doc Bridge](https://doc-bridge.agentskit.io/) — Turn repository documentation into executable agent handoffs.\n- [AgentsKit Code Review](https://code-review.agentskit.io/docs) — Run deep, low-noise review with the model already in use.\n- [AgentsKit Harness](https://harness.agentskit.io/docs) — The keep-pushing loop for your SDLC.\n- [Agents Playbook](https://playbook.agentskit.io/docs) — Make agents ship code a human would actually merge.\n\n> Draft generated by the content pipeline. Human approval required before publishing.\n",
+    "docsPage": "---\ntitle: Connect AgentsKit to your coding agent\ndescription: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n---\n\n# Connect AgentsKit to your coding agent\n\nDevelopers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n## Audience\n\nJavaScript and TypeScript developers using Codex, Claude Code or Desktop, Cursor, Cline, Continue, or another MCP-compatible host\n\n## Commands\n\n```bash\ncodex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search\npnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts\npnpm --filter @agentskit/mcp smoke:published\n```\n\n## Verified claims\n\n- **agentskit/packages**: 22 packages (from ecosystem-claims.json)\n- **agentskit/integrations**: 50 integrations (from ecosystem-claims.json)\n\n## Citations\n\n- [coding-agent MCP host fixture](packages/mcp/fixtures/coding-agent-hosts.ts)\n- [coding-agent MCP recipe](/docs/reference/recipes/coding-agent-mcp)\n- [Codex MCP configuration](https://developers.openai.com/codex/mcp/)\n- [Claude Code MCP configuration](https://docs.anthropic.com/en/docs/claude-code/mcp)\n- [Cursor MCP configuration](https://cursor.com/docs/mcp)\n- [Cline MCP configuration](https://docs.cline.bot/mcp/mcp-overview)\n- [Continue MCP configuration](https://docs.continue.dev/customize/deep-dives/mcp)\n\n## Ecosystem next steps\n\n- [AgentsKit](https://www.agentskit.io/docs) — Build agents without gluing many libraries together.\n- [AgentsKit Registry](https://registry.agentskit.io/docs) — Copy ready-made agents and own the source.\n- [AgentsKit Chat](https://chat.agentskit.io/docs) — Define one agent experience and deliver it across interfaces.\n- [Doc Bridge](https://doc-bridge.agentskit.io/) — Turn repository documentation into executable agent handoffs.\n- [AgentsKit Code Review](https://code-review.agentskit.io/docs) — Run deep, low-noise review with the model already in use.\n- [AgentsKit Harness](https://harness.agentskit.io/docs) — The keep-pushing loop for your SDLC.\n- [Agents Playbook](https://playbook.agentskit.io/docs) — Make agents ship code a human would actually merge.\n\n> Draft generated by the content pipeline. Human approval required before publishing.\n",
     "shortPost": "Connect AgentsKit to your coding agent.\n\n4 commands. No provider key.\n\n```bash\npnpm --filter @agentskit/mcp smoke:published\n```\n\nBuilt on AgentsKit (22 packages, 50 integrations).\n\nDocs + fixture linked in the atom. Human-reviewed drafts only.",
-    "thread": "1/ Connect AgentsKit to your coding agent\n\nWhy it exists: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n2/ The path:\n1. `codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`\n2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`\n3. `pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts`\n4. `pnpm --filter @agentskit/mcp smoke:published`\n\n3/ Claims (generated, not hand-typed):\n- 22 packages\n- 50 integrations\n\n4/ Next products:\n- AgentsKit: Build agents without gluing many libraries together.\n- AgentsKit Registry: Copy ready-made agents and own the source.\n- AgentsKit Chat: Define one agent experience and deliver it across interfaces.\n\n5/ This thread is a draft atom. Publishing requires explicit human approval.",
-    "communityPost": "## Connect AgentsKit to your coding agent\n\nDevelopers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n### Try it\n\n```bash\ncodex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search\npnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts\npnpm --filter @agentskit/mcp smoke:published\n```\n\n### Claims\n22 packages, 50 integrations\n\nSource fixture: `packages/mcp/fixtures/run-coding-agent-hosts.mjs`\n\nStatus: draft — awaiting human approval before community publish.",
+    "thread": "1/ Connect AgentsKit to your coding agent\n\nWhy it exists: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n2/ The path:\n1. `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`\n2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`\n3. `pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts`\n4. `pnpm --filter @agentskit/mcp smoke:published`\n\n3/ Claims (generated, not hand-typed):\n- 22 packages\n- 50 integrations\n\n4/ Next products:\n- AgentsKit: Build agents without gluing many libraries together.\n- AgentsKit Registry: Copy ready-made agents and own the source.\n- AgentsKit Chat: Define one agent experience and deliver it across interfaces.\n\n5/ This thread is a draft atom. Publishing requires explicit human approval.",
+    "communityPost": "## Connect AgentsKit to your coding agent\n\nDevelopers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\n\n### Try it\n\n```bash\ncodex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search\npnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts\npnpm --filter @agentskit/mcp smoke:published\n```\n\n### Claims\n22 packages, 50 integrations\n\nSource fixture: `packages/mcp/fixtures/run-coding-agent-hosts.mjs`\n\nStatus: draft — awaiting human approval before community publish.",
     "example": {
       "kind": "fixture-link",
       "path": "packages/mcp/fixtures/run-coding-agent-hosts.mjs",
       "commands": [
-        "codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
-        "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
+        "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+        "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
         "pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts",
         "pnpm --filter @agentskit/mcp smoke:published"
       ]
     }
   },
-  "storyboard": "# Storyboard — Connect AgentsKit to your coding agent\n\nFormat: 5-frame carousel / 15s silent GIF\n\n1. **Hook** — \"Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\"\n2. **Install** — show `codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`\n3. **Copy fixture** — show the agent.ts path\n4. **Run** — terminal output includes the expected ready line\n5. **Next step** — point to Registry / Playbook / Chat without inventing counts\n\nAccessibility: every frame needs alt text; prefer reduced-motion static frames.\nNo auto-publish — storyboard is a draft for human design review.\n",
+  "storyboard": "# Storyboard — Connect AgentsKit to your coding agent\n\nFormat: 5-frame carousel / 15s silent GIF\n\n1. **Hook** — \"Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.\"\n2. **Install** — show `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`\n3. **Copy fixture** — show the agent.ts path\n4. **Run** — terminal output includes the expected ready line\n5. **Next step** — point to Registry / Playbook / Chat without inventing counts\n\nAccessibility: every frame needs alt text; prefer reduced-motion static frames.\nNo auto-publish — storyboard is a draft for human design review.\n",
   "publish": {
     "status": "blocked",
     "reason": "Human approval required before publish packaging",
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/carousel-storyboard.md b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/carousel-storyboard.md
index 00e880b86..2af18d96c 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/carousel-storyboard.md
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/carousel-storyboard.md
@@ -3,7 +3,7 @@
 Format: 5-frame carousel / 15s silent GIF
 
 1. **Hook** — "Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host."
-2. **Install** — show `codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`
+2. **Install** — show `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`
 3. **Copy fixture** — show the agent.ts path
 4. **Run** — terminal output includes the expected ready line
 5. **Next step** — point to Registry / Playbook / Chat without inventing counts
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md
index 4844f669e..be26916a7 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md
@@ -5,8 +5,8 @@ Developers should be able to use the same AgentsKit tools and agents inside thei
 ### Try it
 
 ```bash
-codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
-claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts
 pnpm --filter @agentskit/mcp smoke:published
 ```
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx
index 6d26d8d27..e6c715bd8 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx
@@ -14,8 +14,8 @@ JavaScript and TypeScript developers using Codex, Claude Code or Desktop, Cursor
 ## Commands
 
 ```bash
-codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
-claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts
 pnpm --filter @agentskit/mcp smoke:published
 ```
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/example.json b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/example.json
index d8f04adc8..8a0ba977d 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/example.json
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/example.json
@@ -2,8 +2,8 @@
   "kind": "fixture-link",
   "path": "packages/mcp/fixtures/run-coding-agent-hosts.mjs",
   "commands": [
-    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
-    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
+    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
     "pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts",
     "pnpm --filter @agentskit/mcp smoke:published"
   ]
diff --git a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/social-thread.md b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/social-thread.md
index 6937fe7bd..94771a557 100644
--- a/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/social-thread.md
+++ b/docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/social-thread.md
@@ -3,8 +3,8 @@
 Why it exists: Developers should be able to use the same AgentsKit tools and agents inside their existing coding agent without building or maintaining a separate integration for every host.
 
 2/ The path:
-1. `codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`
-2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search`
+1. `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`
+2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`
 3. `pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts`
 4. `pnpm --filter @agentskit/mcp smoke:published`
 
diff --git a/docs/ecosystem/content-pipeline/recipes/coding-agent-mcp.json b/docs/ecosystem/content-pipeline/recipes/coding-agent-mcp.json
index 4a9e4e231..499f73581 100644
--- a/docs/ecosystem/content-pipeline/recipes/coding-agent-mcp.json
+++ b/docs/ecosystem/content-pipeline/recipes/coding-agent-mcp.json
@@ -22,8 +22,8 @@
     "expectStdoutIncludes": "verified local MCP stdio protocol; cli tools: fetch_url, web_search"
   },
   "commands": [
-    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
-    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
+    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
     "pnpm --filter @agentskit/mcp exec vitest run tests/coding-agent-hosts.test.ts",
     "pnpm --filter @agentskit/mcp smoke:published"
   ],
diff --git a/packages/adapters/package.json b/packages/adapters/package.json
index 50e0e430d..dd2e71e2d 100644
--- a/packages/adapters/package.json
+++ b/packages/adapters/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/adapters",
-  "version": "0.18.1",
+  "version": "0.19.0",
   "description": "Provider adapters for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/angular/package.json b/packages/angular/package.json
index 54996f626..312341c5d 100644
--- a/packages/angular/package.json
+++ b/packages/angular/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/angular",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "Angular signals service + headless chat components for AgentsKit — streaming, tools, and memory across any LLM provider.",
   "keywords": [
     "agentskit",
diff --git a/packages/cli/package.json b/packages/cli/package.json
index c85812d06..490bf1b7f 100644
--- a/packages/cli/package.json
+++ b/packages/cli/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/cli",
-  "version": "0.13.46",
+  "version": "0.13.47",
   "description": "CLI for AgentsKit chat and project scaffolding.",
   "engines": {
     "node": ">=22"
diff --git a/packages/core/package.json b/packages/core/package.json
index 23726d534..bb3e7ce0d 100644
--- a/packages/core/package.json
+++ b/packages/core/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/core",
-  "version": "1.14.1",
+  "version": "1.15.0",
   "description": "Portable AgentsKit runtime for chat orchestration, tools, memory, and retrieval.",
   "keywords": [
     "agentskit",
diff --git a/packages/cross-platform/package.json b/packages/cross-platform/package.json
index ff573a09c..8f2410875 100644
--- a/packages/cross-platform/package.json
+++ b/packages/cross-platform/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/cross-platform",
-  "version": "0.2.1",
+  "version": "0.3.0",
   "description": "One place for OS and runtime portability in the AgentsKit ecosystem: spawn, paths, line endings and filesystem that behave the same on Windows, macOS and Linux under Node, Bun and Deno.",
   "keywords": [
     "agentskit",
diff --git a/packages/eval-braintrust/package.json b/packages/eval-braintrust/package.json
index 968be3a31..dad7390aa 100644
--- a/packages/eval-braintrust/package.json
+++ b/packages/eval-braintrust/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@agentskit/eval-braintrust",
   "private": true,
-  "version": "0.2.27",
+  "version": "0.2.28",
   "description": "Braintrust scoring pipeline for AgentsKit. Quality and robustness scorers, optional experiment upload, and CI regression alerts.",
   "keywords": [
     "agentskit",
diff --git a/packages/eval/package.json b/packages/eval/package.json
index 170654264..587bb6250 100644
--- a/packages/eval/package.json
+++ b/packages/eval/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/eval",
-  "version": "0.6.12",
+  "version": "0.7.0",
   "description": "Agent evaluation and benchmarking for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/ink/package.json b/packages/ink/package.json
index 8293cedb1..5ee414e4c 100644
--- a/packages/ink/package.json
+++ b/packages/ink/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/ink",
-  "version": "0.10.15",
+  "version": "0.10.16",
   "description": "Ink terminal components and hooks for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/integrations/package.json b/packages/integrations/package.json
index b6027655e..f48b0f91b 100644
--- a/packages/integrations/package.json
+++ b/packages/integrations/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/integrations",
-  "version": "0.8.1",
+  "version": "0.8.2",
   "description": "Unified, plug-and-play service integrations for AgentsKit agents — one descriptor per service projected into tools, connectors, triggers, and auth.",
   "keywords": [
     "agentskit",
diff --git a/packages/mcp/README.md b/packages/mcp/README.md
index 5ccbc9988..42ac0a0e3 100644
--- a/packages/mcp/README.md
+++ b/packages/mcp/README.md
@@ -31,7 +31,7 @@ Docs: [MCP package guide](https://www.agentskit.io/docs/agents/tools/mcp) · [ag
 
 <!-- readme-command:install -->
 ```bash
-npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 ```
 
 Then point your MCP host at the command. Claude Desktop, Cursor, Codex, and
@@ -42,7 +42,7 @@ Claude Desktop or Cursor (`mcpServers` / `.cursor/mcp.json`):
 ```json
 {
   "mcpServers": {
-    "agentskit": { "command": "npx", "args": ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"] }
+    "agentskit": { "command": "npx", "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"] }
   }
 }
 ```
@@ -52,7 +52,7 @@ Codex (`~/.codex/config.toml`):
 ```toml
 [mcp_servers.agentskit]
 command = "npx"
-args = ["-y", "@agentskit/mcp@0.4.10", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
 ```
 
 OpenClaw:
@@ -61,7 +61,7 @@ OpenClaw:
 openclaw mcp add agentskit \
   --command npx \
   --arg -y \
-  --arg @agentskit/mcp@0.4.10 \
+  --arg @agentskit/mcp@0.5.0 \
   --arg --tools \
   --arg fetch,search
 openclaw mcp doctor agentskit --probe
@@ -128,7 +128,7 @@ Run a registry agent server-side and expose it as a single MCP tool — the host
 delegates a specialized job instead of orchestrating primitives:
 
 ```bash
-npx @agentskit/mcp@0.4.10 --agents legal-contract-reviewer,fintech-kyc-screener --provider openai
+npx @agentskit/mcp@0.5.0 --agents legal-contract-reviewer,fintech-kyc-screener --provider openai
 ```
 
 `--provider` covers the first-class adapters (openai, anthropic, gemini, ollama)
diff --git a/packages/mcp/fixtures/coding-agent-hosts.ts b/packages/mcp/fixtures/coding-agent-hosts.ts
index 7450cd89c..106a854c8 100644
--- a/packages/mcp/fixtures/coding-agent-hosts.ts
+++ b/packages/mcp/fixtures/coding-agent-hosts.ts
@@ -3,14 +3,14 @@ import { createInMemoryTransportPair, createMcpClient } from '@agentskit/tools/m
 import { createAgentsKitMcpServer } from '../src/index'
 
 const command = 'npx'
-const args = ['-y', '@agentskit/mcp@0.4.10', '--tools', 'fetch,search'] as const
+const args = ['-y', '@agentskit/mcp@0.5.0', '--tools', 'fetch,search'] as const
 
 const stdioServer = { type: 'stdio', command, args } as const
 const desktopStdioServer = { command, args } as const
 
 export const codingAgentHostConfigs = {
   claude: {
-    cli: 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search',
+    cli: 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search',
     config: {
       mcpServers: {
         agentskit: stdioServer,
@@ -30,7 +30,7 @@ export const codingAgentHostConfigs = {
     },
   },
   codex: {
-    cli: 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search',
+    cli: 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search',
     config: [
       '[mcp_servers.agentskit]',
       `command = "${command}"`,
@@ -63,7 +63,7 @@ export const codingAgentHostConfigs = {
   continue: {
     config: [
       'name: AgentsKit MCP',
-      'version: 0.4.10',
+      'version: 0.5.0',
       'schema: v1',
       'mcpServers:',
       '  - name: agentskit',
@@ -83,9 +83,9 @@ export const codingAgentHostConfigs = {
   },
 } as const
 
-const expectedHostArgs = ['-y', '@agentskit/mcp@0.4.10', '--tools', 'fetch,search']
-const expectedClaudeCli = 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search'
-const expectedCodexCli = 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.4.10 --tools fetch,search'
+const expectedHostArgs = ['-y', '@agentskit/mcp@0.5.0', '--tools', 'fetch,search']
+const expectedClaudeCli = 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search'
+const expectedCodexCli = 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search'
 
 const isRecord = (value: unknown): value is Record<string, unknown> =>
   typeof value === 'object' && value !== null && !Array.isArray(value)
@@ -137,7 +137,7 @@ const validateStdioServer = (
   requireValue(
     Array.isArray(server.args) && server.args.every((arg): arg is string => typeof arg === 'string')
       && JSON.stringify(server.args) === JSON.stringify(expectedHostArgs),
-    `${host}.args must pin @agentskit/mcp@0.4.10 with fetch,search`,
+    `${host}.args must pin @agentskit/mcp@0.5.0 with fetch,search`,
   )
 }
 
@@ -195,7 +195,7 @@ const validateContinueYaml = (source: string): void => {
   const lines = source.split('\n')
   const metadata = [
     'name: AgentsKit MCP',
-    'version: 0.4.10',
+    'version: 0.5.0',
     'schema: v1',
     'mcpServers:',
     '  - name: agentskit',
diff --git a/packages/mcp/fixtures/run-published-coding-agent-hosts.mjs b/packages/mcp/fixtures/run-published-coding-agent-hosts.mjs
index 0d64dfc52..fa3f654d5 100644
--- a/packages/mcp/fixtures/run-published-coding-agent-hosts.mjs
+++ b/packages/mcp/fixtures/run-published-coding-agent-hosts.mjs
@@ -2,7 +2,7 @@ import { spawn } from 'node:child_process'
 import { tmpdir } from 'node:os'
 import { createMcpClient, createStdioTransport } from '@agentskit/tools/mcp'
 
-const packageSpec = '@agentskit/mcp@0.4.10'
+const packageSpec = '@agentskit/mcp@0.5.0'
 const child = spawn('npx', ['-y', packageSpec, '--tools', 'fetch,search'], {
   cwd: tmpdir(),
   stdio: ['pipe', 'pipe', 'pipe'],
diff --git a/packages/mcp/llms-install.md b/packages/mcp/llms-install.md
index d0691a3e8..e3a89ce07 100644
--- a/packages/mcp/llms-install.md
+++ b/packages/mcp/llms-install.md
@@ -4,7 +4,7 @@ Use Node.js 20+ and launch the published server over stdio with the smallest
 default tool set:
 
 ```bash
-npx -y @agentskit/mcp@0.4.10 --tools fetch,search
+npx -y @agentskit/mcp@0.5.0 --tools fetch,search
 ```
 
 The process is an MCP JSON-RPC server. Keep stdout connected to the MCP host;
diff --git a/packages/mcp/package.json b/packages/mcp/package.json
index c5afa8093..15e445b6e 100644
--- a/packages/mcp/package.json
+++ b/packages/mcp/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/mcp",
-  "version": "0.4.10",
+  "version": "0.5.0",
   "description": "Expose AgentsKit tools as an MCP server for hosts supporting the documented 2024-11-05 tools bridge.",
   "keywords": [
     "codex",
diff --git a/packages/mcp/tests/coding-agent-hosts.test.ts b/packages/mcp/tests/coding-agent-hosts.test.ts
index b726a16eb..2e9e88785 100644
--- a/packages/mcp/tests/coding-agent-hosts.test.ts
+++ b/packages/mcp/tests/coding-agent-hosts.test.ts
@@ -5,7 +5,7 @@ import {
   verifyMcpProtocol,
 } from '../fixtures/coding-agent-hosts'
 
-const expectedArgs = ['-y', '@agentskit/mcp@0.4.10', '--tools', 'fetch,search']
+const expectedArgs = ['-y', '@agentskit/mcp@0.5.0', '--tools', 'fetch,search']
 
 describe('coding-agent MCP host recipe', () => {
   it('validates every JSON, TOML, and YAML wrapper', () => {
@@ -101,7 +101,7 @@ describe('coding-agent MCP host recipe', () => {
 
     const driftingCli = JSON.parse(JSON.stringify(codingAgentHostConfigs)) as Record<string, unknown>
     const codex = driftingCli.codex as Record<string, unknown>
-    codex.cli = String(codex.cli).replace('@agentskit/mcp@0.4.10', '@agentskit/mcp')
+    codex.cli = String(codex.cli).replace('@agentskit/mcp@0.5.0', '@agentskit/mcp')
     expect(() => validateCodingAgentHostConfigs(driftingCli)).toThrow(/Codex CLI command must remain version-pinned/)
   })
 
diff --git a/packages/memory/package.json b/packages/memory/package.json
index d0e7f9da4..da3a0cc80 100644
--- a/packages/memory/package.json
+++ b/packages/memory/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/memory",
-  "version": "0.11.14",
+  "version": "0.12.0",
   "description": "Persistent and vector memory backends for AgentsKit.",
   "keywords": [
     "agentskit",
@@ -102,7 +102,7 @@
   },
   "agentskit": {
     "stability": "beta",
-    "stabilityNote": "File/SQLite/Redis backends \u2014 Memory contract v1."
+    "stabilityNote": "File/SQLite/Redis backends — Memory contract v1."
   },
   "license": "MIT",
   "homepage": "https://www.agentskit.io/docs/reference/packages/memory",
diff --git a/packages/net/package.json b/packages/net/package.json
index 05917082e..a9c575df8 100644
--- a/packages/net/package.json
+++ b/packages/net/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/net",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Resilient, bounded and SSRF-safe HTTP for the AgentsKit ecosystem: retry with Retry-After, timeouts, size-limited bodies, SSE parsing and private-address checks. Runs in Node, Bun, Deno, browsers and edge.",
   "keywords": [
     "agentskit",
diff --git a/packages/observability-langfuse/package.json b/packages/observability-langfuse/package.json
index 348faaa56..abcf01cbe 100644
--- a/packages/observability-langfuse/package.json
+++ b/packages/observability-langfuse/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@agentskit/observability-langfuse",
   "private": true,
-  "version": "0.2.26",
+  "version": "0.2.27",
   "description": "Langfuse tracing adapter for AgentsKit. Spans for plan, tool, and model events with token and latency metadata.",
   "keywords": [
     "agentskit",
diff --git a/packages/observability/package.json b/packages/observability/package.json
index 9745fbab6..8eff405b2 100644
--- a/packages/observability/package.json
+++ b/packages/observability/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/observability",
-  "version": "0.12.1",
+  "version": "0.13.0",
   "description": "Logging and tracing layer for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/packages/rag/package.json b/packages/rag/package.json
index 95b83b9ba..203da322c 100644
--- a/packages/rag/package.json
+++ b/packages/rag/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/rag",
-  "version": "0.6.1",
+  "version": "0.7.0",
   "description": "Plug-and-play retrieval-augmented generation for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/react-native/package.json b/packages/react-native/package.json
index 82c00a12c..d6d350abd 100644
--- a/packages/react-native/package.json
+++ b/packages/react-native/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/react-native",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "React Native + Expo hook and headless chat components for AgentsKit — streaming, tools, and memory across any LLM provider.",
   "keywords": [
     "agentskit",
diff --git a/packages/react/package.json b/packages/react/package.json
index ebbb0ea38..c3342b53c 100644
--- a/packages/react/package.json
+++ b/packages/react/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/react",
-  "version": "0.8.9",
+  "version": "0.8.10",
   "description": "React hooks and UI components for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/runtime/package.json b/packages/runtime/package.json
index 05dffba3b..45a4da411 100644
--- a/packages/runtime/package.json
+++ b/packages/runtime/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/runtime",
-  "version": "0.11.1",
+  "version": "0.11.2",
   "description": "Standalone agent runtime with ReAct loop for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/sandbox/package.json b/packages/sandbox/package.json
index efc144b2c..a178f4504 100644
--- a/packages/sandbox/package.json
+++ b/packages/sandbox/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/sandbox",
-  "version": "0.6.12",
+  "version": "0.6.13",
   "description": "Secure code execution for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/packages/skills/package.json b/packages/skills/package.json
index 222ca3435..18cd360c4 100644
--- a/packages/skills/package.json
+++ b/packages/skills/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/skills",
-  "version": "0.9.9",
+  "version": "0.9.10",
   "description": "Ready-made skills (prompts + behavioral instructions) for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/packages/solid/package.json b/packages/solid/package.json
index 47b67deaa..66cea8dc5 100644
--- a/packages/solid/package.json
+++ b/packages/solid/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/solid",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "Solid hook + headless chat components for AgentsKit — streaming, tools, and memory across any LLM provider.",
   "keywords": [
     "agentskit",
diff --git a/packages/statechart/package.json b/packages/statechart/package.json
index 751f3c586..6a777b4c8 100644
--- a/packages/statechart/package.json
+++ b/packages/statechart/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/statechart",
-  "version": "0.3.1",
+  "version": "0.3.2",
   "description": "Deterministic, serializable interaction state for AgentsKit applications.",
   "keywords": [
     "agentskit",
diff --git a/packages/svelte/package.json b/packages/svelte/package.json
index f8848e85a..3ba82e473 100644
--- a/packages/svelte/package.json
+++ b/packages/svelte/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/svelte",
-  "version": "0.5.9",
+  "version": "0.6.0",
   "description": "Svelte 5 store + headless chat components for AgentsKit — streaming, tools, and memory across any LLM provider.",
   "keywords": [
     "agentskit",
diff --git a/packages/templates/package.json b/packages/templates/package.json
index 1f329f400..7a4cbd9f7 100644
--- a/packages/templates/package.json
+++ b/packages/templates/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/templates",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "Extension toolkit for creating custom AgentsKit skills, tools, and adapters.",
   "keywords": [
     "agentskit",
diff --git a/packages/tools/package.json b/packages/tools/package.json
index da26fa084..d10d17c42 100644
--- a/packages/tools/package.json
+++ b/packages/tools/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/tools",
-  "version": "0.13.14",
+  "version": "0.13.15",
   "description": "Reusable executable tools for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/packages/validation/package.json b/packages/validation/package.json
index c4290b645..1a4ac8cc4 100644
--- a/packages/validation/package.json
+++ b/packages/validation/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@agentskit/validation",
   "private": true,
-  "version": "0.2.22",
+  "version": "0.2.23",
   "description": "Opt-in runtime validation of tool-call arguments against their JSON Schema for AgentsKit. Wraps Ajv; plugs into the core ArgsValidator contract (ADR-0008).",
   "keywords": [
     "agentskit",
diff --git a/packages/vue/package.json b/packages/vue/package.json
index 33b4394d0..d9b6dfceb 100644
--- a/packages/vue/package.json
+++ b/packages/vue/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/vue",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "Vue 3 composable + headless chat components for AgentsKit — streaming, tools, and memory across any LLM provider.",
   "keywords": [
     "agentskit",
diff --git a/readme-standard-v1.json b/readme-standard-v1.json
index 050e7474e..7d7db85f0 100644
--- a/readme-standard-v1.json
+++ b/readme-standard-v1.json
@@ -285,7 +285,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:5b20b7ab61f667e9a5801a71d15c6e9ad689bf25a22dc8d264cee669ba13032d"
+        "sourceHash": "sha256:4d0071d088c0467dc2c768b1dda3b146b7921dcb3d3dbd163378135be7353894"
       },
       "exceptions": []
     },
@@ -363,7 +363,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:315772d52e38517441f7c58937840dfe1c57906c93ba0e69e63e477edb85fcf2"
+        "sourceHash": "sha256:9300aad7c0a4174f98873a4337feff10712199a3f3161f98d8d1f04c39706168"
       },
       "exceptions": []
     },
@@ -441,7 +441,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:d92ff745a5316b47b267ee4972c3af28bf9782815b9e51c69872f9de4d2464e1"
+        "sourceHash": "sha256:ea37b973d118aaceeeb4e54b0cc1ff6f6a149c2d9549e5a6e0d72f4793de7c32"
       },
       "exceptions": []
     },
@@ -519,7 +519,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:e6346be6222d7760d6682774b3c775cb14dda438182be499796e8f7609d37c8b"
+        "sourceHash": "sha256:c2cf47169a8dd6523bddfc66a4f60bbf95c524c0035749e4ce65016ac6a3e6d3"
       },
       "exceptions": []
     },
@@ -597,7 +597,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:01b3ea395ab4569875378850827f0b05125d33c5f6fc0927dd7ec8078c403c04"
+        "sourceHash": "sha256:bbe8291c7b04c02dfa3cdd5ebe691b6f20b18069f3e27691b65dbe41389b70dc"
       },
       "exceptions": []
     },
@@ -675,7 +675,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:cd25b6897efef00ec76b11779baf402c67905e1f1ac46622a8a0a118a6e6492d"
+        "sourceHash": "sha256:1948b69de207f398ed87dd6157e4e6fb839eec8dfca44cb76df199ce08e3499e"
       },
       "exceptions": []
     },
@@ -753,7 +753,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:35a7dc80473d6b50ff323f3cae9d352e0b32fb585633e6a5c6fbf63852a3f88b"
+        "sourceHash": "sha256:8da705b7a8f599e2f1bf81c34bca12486e3eaeca7df0ba976eea2eca875dd385"
       },
       "exceptions": []
     },
@@ -831,7 +831,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:3deea003cdd5e6f6f4b2684455faec904b1fda9b69913796a35f23ea36540397"
+        "sourceHash": "sha256:11f5c3016cf7784ac015e8d1d2c36d1116f61a7bfe7ca99bcbea7a2f00f40a1d"
       },
       "exceptions": []
     },
@@ -886,7 +886,7 @@
       "commands": [
         {
           "id": "install",
-          "command": "npx -y @agentskit/mcp@0.4.10 --tools fetch,search",
+          "command": "npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
           "test": "packages/mcp/tests/server.test.ts",
           "testCommand": "pnpm --filter @agentskit/mcp test"
         }
@@ -909,7 +909,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:68fee3474512a719c1d3c5cf2249ef50cb8d47134ef2cba956ab0d0852a2086f"
+        "sourceHash": "sha256:0b8820a1d540c8d160f3e153e911f688936eac2a0701452adc0a2a49a0772acd"
       },
       "exceptions": []
     },
@@ -987,7 +987,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:0820eaa7069c0604769119af5d4c70111cef751f4cbba6f6acfe28d5e0236a10"
+        "sourceHash": "sha256:ff0b8d9c4f06a78b1b38513e11f84bd64ab42ce3048bc66a7b3446383e6b74cd"
       },
       "exceptions": []
     },
@@ -1065,7 +1065,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:6cfb38c8e4527f34e92a201e90c4a96d58468b1301db73511104c699f5ccd2c6"
+        "sourceHash": "sha256:580fb37b52c1f8b4a0b5074ee2dfd007d2265451563ef33acd96896233a888b6"
       },
       "exceptions": []
     },
@@ -1143,7 +1143,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:68fd6bf0c01ab922237f698693aa7fe4781c7f48f0b5cff71054ab453d7c2116"
+        "sourceHash": "sha256:4835c063190d2dcfbf7e983b6cda9ba48c30e1e5b2a7956a691a99f375a9808a"
       },
       "exceptions": []
     },
@@ -1221,7 +1221,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:9c2d77b83a758d76e8dae8d195f1a6721f35acf4577352264eaa9320e6be4bfb"
+        "sourceHash": "sha256:43a21a84b90afafd9d574077ac0f7f5ae183f034af94e42e242dfbf9f5844dc4"
       },
       "exceptions": []
     },
@@ -1299,7 +1299,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:6a206bcb68f7360ebd5ac4ff2b04a238d138887704d5b5f8999c0342682f0b4f"
+        "sourceHash": "sha256:227b156719314891b7eccac5367d94223a30f4140ef58875183c6c821e41191b"
       },
       "exceptions": []
     },
@@ -1377,7 +1377,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:99ab330f64d1cd8afafae6af8b13ecbccbc66b2e0d4742a34f0e35212eac71a5"
+        "sourceHash": "sha256:0c8e930b361095e0de9e5607b2b7777a0aa1e1b42369edc908f04b53647670ce"
       },
       "exceptions": []
     },
@@ -1455,7 +1455,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:3c2dae1ed8069d4ecc41b8b24eae8a4f07ea8b06d12492118712bf73b13e3326"
+        "sourceHash": "sha256:1dc0ae320205343af9a605be5de40877f8c358c7a23747d7de57e822200af4e2"
       },
       "exceptions": []
     },
@@ -1533,7 +1533,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:d9ea5e166673fc35acd436ccb37f2facacded263922709e2588fc806c38f1a5c"
+        "sourceHash": "sha256:b4af0da3964fb1024a9215ccfc9b1a488d211ffb5ad2f9b98edaf5ad809a82f9"
       },
       "exceptions": []
     },
@@ -1611,7 +1611,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:70dc1b6f830675c3fd17e75b7e45bc7ec96ebcec75ab6a974fc0891ef98b52e5"
+        "sourceHash": "sha256:49054391c94bf3d8ed9938fa9e0a97d37dcdcdd5075cc094f3c1d470ca0df57b"
       },
       "exceptions": []
     },
@@ -1689,7 +1689,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:e1f03d8c79baff921c623eb11c18077f8be4b2d86fe5dbd08d561980a7d5182f"
+        "sourceHash": "sha256:93458743b380ac5d62486a1410a3793762ed3fc2e5a17d9d1f8080b0fa994a8b"
       },
       "exceptions": []
     },
@@ -1767,7 +1767,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:f94b44763c8edae7b120a26af3e0e150c885e091aeea9a092f2029b51297e94f"
+        "sourceHash": "sha256:2fe13e8c3cd23cf13c55dbf46da7219f77965419c5d30b69c4a87777abc9f258"
       },
       "exceptions": []
     },
@@ -1845,7 +1845,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:0df203a334c4d97d0c5e4d8f94fad3f99e5a4b8384e55057821e51cb1e72b40a"
+        "sourceHash": "sha256:c6c5313f0f773474baf1825f2156e7c0a2640092b957879d24dc0eab69d29d92"
       },
       "exceptions": []
     },
@@ -1923,7 +1923,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:532cfaa4b31709381ad0b9eaac3dcb36df37f3cb8b7ec6c03a403a3c6429a019"
+        "sourceHash": "sha256:e1c3a47a84245ee55a4352c3ff6f7e9b18a253977b6f6aab8af191b9168a6c0c"
       },
       "exceptions": []
     },
@@ -2001,7 +2001,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:84b08546bb21ee36cd362d8962f40f8204604508ae82b5f9f18f31aa839d54e0"
+        "sourceHash": "sha256:39c26869906b8ea7b0f6076ea4fdb142ddb0171b02bcbeb3d305d8ca3db965a9"
       },
       "exceptions": []
     },
@@ -2079,7 +2079,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:bc6a30bfc11e260c4e086976d449e50677bd6d65f0324590b60e33d298d73e20"
+        "sourceHash": "sha256:2406ce245ae44ec7010813ea909a1327baa0cbdb63fbf0cb665d78a318ce6ed9"
       },
       "exceptions": []
     },
@@ -2157,7 +2157,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:b4b61ffb7662d60eb776d45aba7f82c39c854f13a1b29708c54c535c1594dd78"
+        "sourceHash": "sha256:7d35da0e7a7c73d6f06a2688d99bfdf070812b2310e91e8677ea02f193f199db"
       },
       "exceptions": []
     },
@@ -2235,7 +2235,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:d54ee93adefc494e85741a14034590c99c184a1758b61c08e4ca02efac3f4383"
+        "sourceHash": "sha256:ddac6b1aa51831f35279044793493f9511f5061f5f159b52ba2e8e861a4f3c29"
       },
       "exceptions": []
     },
@@ -2313,7 +2313,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:1b819dd5d4506e58fa71400837cd8eb2e950f98940a944e0547682056c5fda43"
+        "sourceHash": "sha256:ba9adf693c42bea6d5f1a5bea42ad640bb54fda8620e78cc441357213e5d06d3"
       },
       "exceptions": []
     },