From 6e2ae0b4d3ac37e6fba9d1ef4dd50b50ebfb32a2 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Sat, 24 Oct 2026 02:23:45 +0000
Subject: [PATCH] chore: version packages

---
 .changeset/third-party-minor-refresh.md       |   6 -
 .doc-bridge/capabilities.json                 |   2 +-
 .doc-bridge/index.json                        | 146 +++++++++---------
 .../docs-next/content/docs/for-agents/mcp.mdx |   4 +-
 .../docs/reference/packages/adapters.mdx      |   2 +-
 .../content/docs/reference/packages/cli.mdx   |   2 +-
 .../docs/reference/packages/memory.mdx        |   2 +-
 .../content/docs/reference/packages/net.mdx   |   2 +-
 .../packages/observability-langfuse.mdx       |   2 +-
 .../docs/reference/packages/observability.mdx |   2 +-
 .../content/docs/reference/packages/rag.mdx   |   2 +-
 .../docs/reference/packages/runtime.mdx       |   2 +-
 .../content/docs/reference/packages/tools.mdx |   2 +-
 .../reference/recipes/coding-agent-mcp.mdx    |  20 +--
 .../docs/reference/recipes/mcp-bridge.mdx     |   8 +-
 .../atoms/coding-agent-mcp/atom.json          |  16 +-
 .../coding-agent-mcp/carousel-storyboard.md   |   2 +-
 .../atoms/coding-agent-mcp/community-post.md  |   4 +-
 .../atoms/coding-agent-mcp/docs.mdx           |   4 +-
 .../atoms/coding-agent-mcp/example.json       |   4 +-
 .../atoms/coding-agent-mcp/social-thread.md   |   4 +-
 .../recipes/coding-agent-mcp.json             |   4 +-
 packages/adapters/package.json                |   2 +-
 packages/cli/package.json                     |   2 +-
 packages/integrations/package.json            |   2 +-
 packages/mcp/README.md                        |  10 +-
 packages/mcp/fixtures/coding-agent-hosts.ts   |  18 +--
 .../run-published-coding-agent-hosts.mjs      |   2 +-
 packages/mcp/llms-install.md                  |   2 +-
 packages/mcp/package.json                     |   2 +-
 packages/mcp/tests/coding-agent-hosts.test.ts |   4 +-
 packages/memory/package.json                  |   2 +-
 packages/net/package.json                     |   2 +-
 packages/observability-langfuse/package.json  |   2 +-
 packages/observability/package.json           |   2 +-
 packages/rag/package.json                     |   2 +-
 packages/runtime/package.json                 |   2 +-
 packages/tools/package.json                   |   2 +-
 readme-standard-v1.json                       |  24 +--
 39 files changed, 159 insertions(+), 165 deletions(-)
 delete mode 100644 .changeset/third-party-minor-refresh.md

diff --git a/.changeset/third-party-minor-refresh.md b/.changeset/third-party-minor-refresh.md
deleted file mode 100644
index eb201f07b..000000000
--- a/.changeset/third-party-minor-refresh.md
+++ /dev/null
@@ -1,6 +0,0 @@
----
-"@agentskit/cli": patch
-"@agentskit/net": patch
----
-
-Raise third-party runtime dependency floors to their newest in-major releases: `@inquirer/prompts` 8.7.4, `react` 19.3.1 and `tsx` 4.23.15 in the CLI, and `eventsource-parser` 4.1.2 in net.
diff --git a/.doc-bridge/capabilities.json b/.doc-bridge/capabilities.json
index 7843fe2f9..eda743bd1 100644
--- a/.doc-bridge/capabilities.json
+++ b/.doc-bridge/capabilities.json
@@ -7,7 +7,7 @@
     "name": "AgentsKit",
     "root": "."
   },
-  "contentHash": "e30c41afd86eef2cbd264342d8925c4531db6df699a1bccf023998fa4e496bf9",
+  "contentHash": "dc9fe5e9beded7091d00c6a2e99f0629bd2f706d91a5fad443e20ef5a4e42e24",
   "artifacts": {
     "index": ".doc-bridge/index.json",
     "llmsTxt": ".doc-bridge/llms.txt"
diff --git a/.doc-bridge/index.json b/.doc-bridge/index.json
index d729b5191..85a141a75 100644
--- a/.doc-bridge/index.json
+++ b/.doc-bridge/index.json
@@ -1,8 +1,8 @@
 {
   "schemaVersion": 1,
-  "contentHash": "e30c41afd86eef2cbd264342d8925c4531db6df699a1bccf023998fa4e496bf9",
+  "contentHash": "dc9fe5e9beded7091d00c6a2e99f0629bd2f706d91a5fad443e20ef5a4e42e24",
   "contentHashAlgo": "sha256-semantic-v1",
-  "generatedAt": "2026-10-24T02:03:21.645Z",
+  "generatedAt": "2026-10-24T02:23:43.693Z",
   "project": {
     "name": "AgentsKit",
     "root": "."
@@ -3687,7 +3687,7 @@
         "document",
         "human"
       ],
-      "contentHash": "2f8e5695adf8368b45b9e2fff381159ba18324b21e29ae8777b459030f194353"
+      "contentHash": "3ff4b8ff39bcce16a904a132c116f1d30fc899dde6fae02000f8b1379e168c80"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/angular.mdx",
@@ -3711,7 +3711,7 @@
         "document",
         "human"
       ],
-      "contentHash": "89b93eb9ebacbd746030f3f819fd82f986946bd64cd54c8786ec4f94c24a6cc8"
+      "contentHash": "f728d0c891342cd8be36f3400d774933e6ad73070a7f19791e0993a7e7441ec9"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/core.mdx",
@@ -3783,7 +3783,7 @@
         "document",
         "human"
       ],
-      "contentHash": "8dd9f05860de712e131eb3c96eae190716c20212f9c3595023836729b5eb90bd"
+      "contentHash": "c38e2c7867991813644ed8f948c82d308cd9e72413493f79c50647926ca9a624"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/net.mdx",
@@ -3795,19 +3795,19 @@
         "document",
         "human"
       ],
-      "contentHash": "c5c16474c2949034eb3ba0c226abf9dd5661633975de783a245b81e3d271b2b9"
+      "contentHash": "3511fc16cc0cd811b85d6032c85885b12163ae59f54b3fc74f8068865a6bda60"
     },
     {
       "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: 0.2.27",
+      "description": "Version: 0.2.28",
       "tags": [
         "document",
         "human"
       ],
-      "contentHash": "acb7a721d0c3fea44e57fa429d9778edd8e7312a82b37e5b79599a180416bdb0"
+      "contentHash": "b7cc04cd8a7a9cbd643a10dfbcea69eb731845e64f0fc07f980956b75caa1d5c"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/observability.mdx",
@@ -3819,7 +3819,7 @@
         "document",
         "human"
       ],
-      "contentHash": "4daf7c1c8db1aafdf83d90c55dee29c94d5172669877f29c66dc241df1aafa0f"
+      "contentHash": "cc52aa8f0dd83136c38abcd0ff53f5e0e20c15a096e988f2c0d3628b168a4d48"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/overview.mdx",
@@ -3843,7 +3843,7 @@
         "document",
         "human"
       ],
-      "contentHash": "ef350a62dd4e53a8a700209289ece498cad7ec12adec467e9c47da3d86611a7f"
+      "contentHash": "7ef94859ed43a537d12e477b7b86c45ebe6c928cb4926df152a0e0946992e65e"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/react-native.mdx",
@@ -3891,7 +3891,7 @@
         "document",
         "human"
       ],
-      "contentHash": "e036a7f1a0e0d59200f2d478a727021591159b71b63373ae9fbd2072dfb5dbef"
+      "contentHash": "8069eac1304cfbd1bc9515b0e66762006a0b768f6ad72162b199ec15f5243b52"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/sandbox.mdx",
@@ -3975,7 +3975,7 @@
         "document",
         "human"
       ],
-      "contentHash": "f191d6d99e58f14104dcb644021c464391703a59bdc726c7f47f02681d60ada2"
+      "contentHash": "9fc21430532f38057801de4bbc8e19dc5a4eedbd3a8ae36709285098be2448f3"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/packages/vue.mdx",
@@ -4131,7 +4131,7 @@
         "document",
         "human"
       ],
-      "contentHash": "6c0deafff5744176d351b6c42aae19daab54983b5e6f1ee286ec8d0547ca21b5"
+      "contentHash": "ee1dcbe26fcf142d8aa62f32e0e032217afc0786f937ffdaf42897907999df18"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/recipes/confirmation-gated-tool.mdx",
@@ -4395,7 +4395,7 @@
         "document",
         "human"
       ],
-      "contentHash": "78128080568b626ffc2fa51aecb47bd4117340555df3724668407d9727ad5123"
+      "contentHash": "87e367e28522fd2c6d7a7bd0716742f568605c00e9309cfebc627a247a3ac2c4"
     },
     {
       "id": "document:apps/docs-next/content/docs/reference/recipes/more-integrations.mdx",
@@ -6158,7 +6158,7 @@
         "document",
         "human"
       ],
-      "contentHash": "98a26068fa0636dd6d79762b1b3606616e5b823f49ee4f86718f147446fdf8fd"
+      "contentHash": "fbf1841972e46865b886fe0aa039ad81f492678c35e313a793cf9f52a7aaacb5"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/community-post.md",
@@ -6170,7 +6170,7 @@
         "document",
         "human"
       ],
-      "contentHash": "90e92591b6bc6b4b991729c7eda9deb21e37bbe5ff482f40c713691e82438c4d"
+      "contentHash": "97001ba4b10d435b1a03a1ed1bb4332b3e414ea6a70797c4830f0da6aba4d006"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/docs.mdx",
@@ -6182,7 +6182,7 @@
         "document",
         "human"
       ],
-      "contentHash": "0c786965cab639ef1728feb7b4355039740f699811aaac3bd529a33e9ea89fc6"
+      "contentHash": "365844b46e9f099f336b615e0f97a8bbd479e49c6f235de8952ff1f5387ef4ab"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/coding-agent-mcp/review.md",
@@ -6218,7 +6218,7 @@
         "document",
         "human"
       ],
-      "contentHash": "05e7837ff75b4b53c65053d8843d6d575ba35bcc12d4ee157c5d5d04cd6b9310"
+      "contentHash": "133be1d53d4e9e4d3e472a06b74bb8e42c8d1002c684ab47091bb5c2ee42f7dd"
     },
     {
       "id": "document:docs/ecosystem/content-pipeline/atoms/first-agent/carousel-storyboard.md",
@@ -7202,7 +7202,7 @@
         "document",
         "unclassified"
       ],
-      "contentHash": "4306e774390c1c87a40c6cb508eab760bdb0886c2d0b5ac5ce0180e90ba9c574"
+      "contentHash": "14297baf3a07ff4d8347117b0d7d415dfb6af26fe5797716c5299676b40cfb23"
     },
     {
       "id": "document:packages/mcp/README.md",
@@ -7214,7 +7214,7 @@
         "document",
         "project"
       ],
-      "contentHash": "3097341299c6313b1a1b6cfafddb2149faf3137477da3d953f6a9495dc8cd8f7"
+      "contentHash": "ef58a7a4a8b3ce1d07d1feb14502cae2dce5126e4e4973d7a0b1c0f1a424ee03"
     },
     {
       "id": "document:packages/memory/CONVENTIONS.md",
@@ -25831,7 +25831,7 @@
         "ts",
         "fixtures"
       ],
-      "contentHash": "999745c52fe403d21ecd122d746b8f48f0c519ed557c4f9a58d19a52341d729f"
+      "contentHash": "74f23a8b95de178f9bb21f4ab5f5c2215b55a0f2424ecceb4031abd68abdce5a"
     },
     {
       "id": "module:packages/mcp/fixtures/readme-quickstart.ts",
@@ -25867,7 +25867,7 @@
         "mjs",
         "fixtures"
       ],
-      "contentHash": "7d404d08513159d609e6039a3ed49c4a0a0294d9d629515d51d3dbdcadb989d7"
+      "contentHash": "afa66632292bba93cde0d67fcdc5345b724c3bf658a831789ca170be4c599438"
     },
     {
       "id": "module:packages/mcp/src/agent-tool.ts",
@@ -26011,7 +26011,7 @@
         "test",
         "tests"
       ],
-      "contentHash": "090e00a1a759cff1439b2606e7571e19d27cd6f92f9d20284d712caeb8f7f363"
+      "contentHash": "ef16725b19a0c44869285ca1470147612427bed3f3ba439a6833dc3982a03f3b"
     },
     {
       "id": "module:packages/mcp/tests/package-manifest.test.ts",
@@ -38697,7 +38697,7 @@
     "changes": {}
   },
   "inputs": {
-    "hash": "817587e877cc907fb619b43179c17c08274cd7905bd6a8a4ea23dcd05c2d3c56",
+    "hash": "fa4605bea9f4a23139d346b229cbabf4cc9ab54384c273727d1cd4457486bb72",
     "fileCount": 2656,
     "projectionVersion": 2
   },
@@ -38720,9 +38720,9 @@
   "projection": {
     "type": "retrieval-index",
     "schemaVersion": 1,
-    "contentHash": "b1474b3aad56ea02405d208f54ba165bc08a6c32582646ff4d635f3eab511fbd",
+    "contentHash": "9536373ff31af839d7fccf0726dfeb4637536faeaddca84f4121301096dc2026",
     "contentHashAlgo": "sha256-semantic-v1",
-    "snapshotHash": "88fcaec057eb804864e0e79935e968d5f79c328918f6b0ac334eccbae9776ab3",
+    "snapshotHash": "28866690406a74f4fb58bad1cbf5416231e0d9f4ab85dfa83d6d9cf80cad0e1a",
     "overlayHash": "f99666ff9119084dd25c2974ae64eb6935ec2aa785ff5ad478fb9d541f7a5e94",
     "configurationHash": "d80c82274083d35277566814974f568159fa9b1442b2a41ee0931a2620d4f4a3",
     "lexiconVersion": 1,
@@ -52288,7 +52288,7 @@
           "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@0.5.0 --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",
+          "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@0.5.1 --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": {
@@ -52350,7 +52350,7 @@
             }
           ]
         },
-        "contentHash": "ff7095d5548807583af0d9a1211cb01ff9e4104e4fa2adb99eceda934c988c87",
+        "contentHash": "b844382873964b2cb7cc43b5905069be77bc478d00db8072e8d73fcbfcfb598f",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -59949,7 +59949,7 @@
           "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: 0.19.0 - 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: 0.19.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60166,7 +60166,7 @@
             }
           ]
         },
-        "contentHash": "2f8e5695adf8368b45b9e2fff381159ba18324b21e29ae8777b459030f194353",
+        "contentHash": "3ff4b8ff39bcce16a904a132c116f1d30fc899dde6fae02000f8b1379e168c80",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60247,7 +60247,7 @@
           "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: 0.13.47 - 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: 0.13.48 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60294,7 +60294,7 @@
             }
           ]
         },
-        "contentHash": "89b93eb9ebacbd746030f3f819fd82f986946bd64cd54c8786ec4f94c24a6cc8",
+        "contentHash": "f728d0c891342cd8be36f3400d774933e6ad73070a7f19791e0993a7e7441ec9",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60776,7 +60776,7 @@
           "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: 0.12.0 - 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: 0.12.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -60888,7 +60888,7 @@
             }
           ]
         },
-        "contentHash": "8dd9f05860de712e131eb3c96eae190716c20212f9c3595023836729b5eb90bd",
+        "contentHash": "c38e2c7867991813644ed8f948c82d308cd9e72413493f79c50647926ca9a624",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60910,7 +60910,7 @@
           "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: 0.2.3 - 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: 0.2.4 - Tier: beta - Contract: evolving Deep dives - — full export list. - Decision record: ADR-0037 ( docs/architecture/adrs/0037-net-package.md ).",
           "aliases": "document human"
         },
         "graph": {
@@ -60942,7 +60942,7 @@
             }
           ]
         },
-        "contentHash": "c5c16474c2949034eb3ba0c226abf9dd5661633975de783a245b81e3d271b2b9",
+        "contentHash": "3511fc16cc0cd811b85d6032c85885b12163ae59f54b3fc74f8068865a6bda60",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -60951,7 +60951,7 @@
         "kind": "document",
         "path": "apps/docs-next/content/docs/reference/packages/observability-langfuse.mdx",
         "title": "observability-langfuse.mdx",
-        "summary": "Version: 0.2.27",
+        "summary": "Version: 0.2.28",
         "audience": "human",
         "aliases": [],
         "tags": [
@@ -60963,8 +60963,8 @@
           "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: 0.2.27",
-          "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:",
+          "summary": "Version: 0.2.28",
+          "body": "- Version: 0.2.28 @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": {
@@ -61026,7 +61026,7 @@
             }
           ]
         },
-        "contentHash": "acb7a721d0c3fea44e57fa429d9778edd8e7312a82b37e5b79599a180416bdb0",
+        "contentHash": "b7cc04cd8a7a9cbd643a10dfbcea69eb731845e64f0fc07f980956b75caa1d5c",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61048,7 +61048,7 @@
           "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: 0.13.0 - 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: 0.13.1 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61145,7 +61145,7 @@
             }
           ]
         },
-        "contentHash": "4daf7c1c8db1aafdf83d90c55dee29c94d5172669877f29c66dc241df1aafa0f",
+        "contentHash": "cc52aa8f0dd83136c38abcd0ff53f5e0e20c15a096e988f2c0d3628b168a4d48",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61346,7 +61346,7 @@
           "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: 0.7.0 - 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.",
+          "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: 0.7.1 - 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.",
           "aliases": "document human"
         },
         "graph": {
@@ -61408,7 +61408,7 @@
             }
           ]
         },
-        "contentHash": "ef350a62dd4e53a8a700209289ece498cad7ec12adec467e9c47da3d86611a7f",
+        "contentHash": "7ef94859ed43a537d12e477b7b86c45ebe6c928cb4926df152a0e0946992e65e",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -61697,7 +61697,7 @@
           "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: 0.11.2 - 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: 0.11.3 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -61764,7 +61764,7 @@
             }
           ]
         },
-        "contentHash": "e036a7f1a0e0d59200f2d478a727021591159b71b63373ae9fbd2072dfb5dbef",
+        "contentHash": "8069eac1304cfbd1bc9515b0e66762006a0b768f6ad72162b199ec15f5243b52",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -62190,7 +62190,7 @@
           "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: 0.13.15 - 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: 0.13.16 - Tier: beta - Contract: evolving - Roadmap: see for what this package needs to reach v1.0. Related - - - - Source npm: · repo:",
           "aliases": "document human"
         },
         "graph": {
@@ -62387,7 +62387,7 @@
             }
           ]
         },
-        "contentHash": "f191d6d99e58f14104dcb644021c464391703a59bdc726c7f47f02681d60ada2",
+        "contentHash": "9fc21430532f38057801de4bbc8e19dc5a4eedbd3a8ae36709285098be2448f3",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -63057,7 +63057,7 @@
             }
           ]
         },
-        "contentHash": "6c0deafff5744176d351b6c42aae19daab54983b5e6f1ee286ec8d0547ca21b5",
+        "contentHash": "ee1dcbe26fcf142d8aa62f32e0e032217afc0786f937ffdaf42897907999df18",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -64066,7 +64066,7 @@
             }
           ]
         },
-        "contentHash": "78128080568b626ffc2fa51aecb47bd4117340555df3724668407d9727ad5123",
+        "contentHash": "87e367e28522fd2c6d7a7bd0716742f568605c00e9309cfebc627a247a3ac2c4",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73081,7 +73081,7 @@
           "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@0.5.0 --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.",
+          "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@0.5.1 --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": {
@@ -73092,7 +73092,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "98a26068fa0636dd6d79762b1b3606616e5b823f49ee4f86718f147446fdf8fd",
+        "contentHash": "fbf1841972e46865b886fe0aa039ad81f492678c35e313a793cf9f52a7aaacb5",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73131,7 +73131,7 @@
             }
           ]
         },
-        "contentHash": "90e92591b6bc6b4b991729c7eda9deb21e37bbe5ff482f40c713691e82438c4d",
+        "contentHash": "97001ba4b10d435b1a03a1ed1bb4332b3e414ea6a70797c4830f0da6aba4d006",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73164,7 +73164,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "0c786965cab639ef1728feb7b4355039740f699811aaac3bd529a33e9ea89fc6",
+        "contentHash": "365844b46e9f099f336b615e0f97a8bbd479e49c6f235de8952ff1f5387ef4ab",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -73252,7 +73252,7 @@
           "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@0.5.0 --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.",
+          "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@0.5.1 --tools fetch,search 2. claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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": {
@@ -73263,7 +73263,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "05e7837ff75b4b53c65053d8843d6d575ba35bcc12d4ee157c5d5d04cd6b9310",
+        "contentHash": "133be1d53d4e9e4d3e472a06b74bb8e42c8d1002c684ab47091bb5c2ee42f7dd",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -79808,7 +79808,7 @@
             }
           ]
         },
-        "contentHash": "4306e774390c1c87a40c6cb508eab760bdb0886c2d0b5ac5ce0180e90ba9c574",
+        "contentHash": "14297baf3a07ff4d8347117b0d7d415dfb6af26fe5797716c5299676b40cfb23",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -79902,7 +79902,7 @@
             }
           ]
         },
-        "contentHash": "3097341299c6313b1a1b6cfafddb2149faf3137477da3d953f6a9495dc8cd8f7",
+        "contentHash": "ef58a7a4a8b3ce1d07d1feb14502cae2dce5126e4e4973d7a0b1c0f1a424ee03",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -150744,7 +150744,7 @@
             }
           ]
         },
-        "contentHash": "999745c52fe403d21ecd122d746b8f48f0c519ed557c4f9a58d19a52341d729f",
+        "contentHash": "74f23a8b95de178f9bb21f4ab5f5c2215b55a0f2424ecceb4031abd68abdce5a",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -150851,7 +150851,7 @@
           "inbound": [],
           "outbound": []
         },
-        "contentHash": "7d404d08513159d609e6039a3ed49c4a0a0294d9d629515d51d3dbdcadb989d7",
+        "contentHash": "afa66632292bba93cde0d67fcdc5345b724c3bf658a831789ca170be4c599438",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -151343,7 +151343,7 @@
             }
           ]
         },
-        "contentHash": "090e00a1a759cff1439b2606e7571e19d27cd6f92f9d20284d712caeb8f7f363",
+        "contentHash": "ef16725b19a0c44869285ca1470147612427bed3f3ba439a6833dc3982a03f3b",
         "provenance": "observed",
         "confidence": "observed"
       },
@@ -193750,7 +193750,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "ced1c43bf34b5dc01d33a96aab699ab70ad193a080d8541f67e0e7750d857415",
+        "contentHash": "7b52a55ab78371e1ff29f05049a71b87c58875818b7c1961e6fa62dca2c7f1a9",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "adapters"
@@ -194222,7 +194222,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "62b1780538de80ddb70805f367695ed3c0da8824f333da20b2c043c7ab0510ca",
+        "contentHash": "3523aae5d39c039b74fcf3f7beddfb933377e900405dd780ba7bac9ec369b583",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "cli"
@@ -195679,7 +195679,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "4d974a2a16448299073c5b6ec373daf36822992839566d0071917982853e2f1f",
+        "contentHash": "4e429d391602daeb9ae0d0cc83fb4723c892d04e41c5fa719690cddb0426e52d",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "integrations"
@@ -195705,7 +195705,7 @@
           "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@0.5.0 --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",
+          "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@0.5.1 --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": {
@@ -195840,7 +195840,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "d5c9d91d26e410f02ea80d653ef98aaf391b7faf280a1a3cc4dfeddfc76a6b27",
+        "contentHash": "f8047fddd68ac5c998366ccc233b14f8a1155438a349e2e681bf98952b8e8ae2",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "mcp"
@@ -196265,7 +196265,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "bb083cc8471b36128076fd6b3c2fceed5c565b7674ea50b4b2f8b55bd8f892e3",
+        "contentHash": "f34168846ba12dc064405ffce4f3232b858d69f1b728192827cba12ba808f280",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "memory"
@@ -196420,7 +196420,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "d0117b0023201a8e6aa43a7a9c8cb34b1e1c06e3dfd7c1fb521d22f3f8661f8d",
+        "contentHash": "3e6d5878681a72ff67d628447aaa8148ee96c4c8fec8582e51f8dc908b4f64eb",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "net"
@@ -196785,7 +196785,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "a70b5a62ff896ddb73090aec2215d6fdaf91d35cd6f2a34f9bb757a2e5bea740",
+        "contentHash": "11b44e155869857828d6b919e51d22ab6bc36e47dec6ce452a3d540909582a0a",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "observability"
@@ -196874,7 +196874,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "6abebcf72c84895160795a49c5f584111721cd15373961d02e884c02aeaab679",
+        "contentHash": "e0d0e704c2064c0c4d8a3efde8e736bc8fbddae5035a3659a8554c205765d40f",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "observability-langfuse"
@@ -197221,7 +197221,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "90c80c8d2ad1153be111a86b932639a69895911d3be037c961c574ee27349eda",
+        "contentHash": "cb6cea5f305d3e673ca303e66b539e8abc1d8c91e29728e2ffbbe6b658bdb154",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "rag"
@@ -198328,7 +198328,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "7424f79791670914dcf1171590f9ffaafcd3c8e5e8ef4664d6e4226419010f19",
+        "contentHash": "b2e8dfdf8d384ef6de553c90451359952f3164cabbda0b08872c83bc8174c55a",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "runtime"
@@ -200067,7 +200067,7 @@
           ],
           "outbound": []
         },
-        "contentHash": "ab90aa32eca8167e3c1606bc842abe02a6497f1469f674955748895fd7cb790c",
+        "contentHash": "231037b435459a5e8c85e18858201389e20b0ac8545ea47b6265a83f619ddb9f",
         "provenance": "observed",
         "confidence": "observed",
         "ownershipId": "tools"
diff --git a/apps/docs-next/content/docs/for-agents/mcp.mdx b/apps/docs-next/content/docs/for-agents/mcp.mdx
index 5d1cd663c..823daf4ab 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@0.5.0
+npm install @agentskit/mcp@0.5.1
 ```
 
 ## Primary exports
@@ -100,7 +100,7 @@ other features from newer protocol revisions.
 
 ## Bin
 
-`agentskit-mcp` — `npx -y @agentskit/mcp@0.5.0 --tools fetch,search`. Expose whole agents
+`agentskit-mcp` — `npx -y @agentskit/mcp@0.5.1 --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 69fae8b44..d818218d8 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:** `0.19.0`
+- **Version:** `0.19.1`
 - **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 8e6c15071..1b5b2b2d1 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:** `0.13.47`
+- **Version:** `0.13.48`
 - **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 eab91045e..d6ef322dc 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:** `0.12.0`
+- **Version:** `0.12.1`
 - **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 066736abd..7150dacc1 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:** `0.2.3`
+- **Version:** `0.2.4`
 - **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 72bbdc785..0b576e49e 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:** `0.2.27`
+- **Version:** `0.2.28`
 
 `@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 d88ac26cf..3f48477ee 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:** `0.13.0`
+- **Version:** `0.13.1`
 - **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 348c6fc22..33d76b27b 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:** `0.7.0`
+- **Version:** `0.7.1`
 - **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 879948a33..fa0f88729 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:** `0.11.2`
+- **Version:** `0.11.3`
 - **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/tools.mdx b/apps/docs-next/content/docs/reference/packages/tools.mdx
index 3a574e260..25e62e6a3 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:** `0.13.15`
+- **Version:** `0.13.16`
 - **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/recipes/coding-agent-mcp.mdx b/apps/docs-next/content/docs/reference/recipes/coding-agent-mcp.mdx
index 99e75ae57..99c34e9f1 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@0.5.0 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.0", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.1", "--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@0.5.0 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search
 claude mcp list
 ```
 
@@ -68,7 +68,7 @@ The equivalent `.mcp.json` is:
     "agentskit": {
       "type": "stdio",
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"]
     }
   }
 }
@@ -89,7 +89,7 @@ Create `.cursor/mcp.json`:
     "agentskit": {
       "type": "stdio",
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"]
     }
   }
 }
@@ -109,7 +109,7 @@ edits `%APPDATA%\Claude\claude_desktop_config.json`.
   "mcpServers": {
     "agentskit": {
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"]
     }
   }
 }
@@ -128,7 +128,7 @@ Open MCP Servers, choose Configure MCP Servers, and add:
   "mcpServers": {
     "agentskit": {
       "command": "npx",
-      "args": ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"],
+      "args": ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"],
       "env": {},
       "disabled": false,
       "autoApprove": []
@@ -151,7 +151,7 @@ Create `.continue/mcpServers/agentskit.yaml`:
 
 ```yaml
 name: AgentsKit MCP
-version: 0.5.0
+version: 0.5.1
 schema: v1
 mcpServers:
   - name: agentskit
@@ -159,7 +159,7 @@ mcpServers:
     command: npx
     args:
       - "-y"
-      - "@agentskit/mcp@0.5.0"
+      - "@agentskit/mcp@0.5.1"
       - "--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@0.5.0 \
+  npx -y @agentskit/mcp@0.5.1 \
   --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 d27b657cb..7afeec836 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@0.5.0 --tools fetch,search
+npx -y @agentskit/mcp@0.5.1 --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@0.5.0", "--tools", "fetch,search"]
+      "args": ["-y", "@agentskit/mcp@0.5.1", "--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@0.5.0", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.1", "--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@0.5.0 \
+  --arg @agentskit/mcp@0.5.1 \
   --arg --tools \
   --arg fetch,search
 openclaw mcp doctor agentskit --probe
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 13b9aeae3..174673912 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@0.5.0 --tools fetch,search",
-      "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+      "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search",
+      "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.0 --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",
+    "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@0.5.1 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.0 --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.",
+    "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@0.5.1 --tools fetch,search`\n2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.1 --tools fetch,search\nclaude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.0 --tools fetch,search",
-        "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+        "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search",
+        "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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@0.5.0 --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",
+  "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@0.5.1 --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 2af18d96c..705e9b972 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@0.5.0 --tools fetch,search`
+2. **Install** — show `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --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 be26916a7..5d9325766 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@0.5.0 --tools fetch,search
-claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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 e6c715bd8..9404cdaa1 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@0.5.0 --tools fetch,search
-claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search
+codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search
+claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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 8a0ba977d..0616f69b1 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@0.5.0 --tools fetch,search",
-    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search",
+    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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 94771a557..20130f219 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@0.5.0 --tools fetch,search`
-2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search`
+1. `codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search`
+2. `claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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 499f73581..c88bad3aa 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@0.5.0 --tools fetch,search",
-    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+    "codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search",
+    "claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --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 f5c8f4625..dbbefe3fd 100644
--- a/packages/adapters/package.json
+++ b/packages/adapters/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/adapters",
-  "version": "0.19.0",
+  "version": "0.19.1",
   "description": "Provider adapters for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/cli/package.json b/packages/cli/package.json
index 32546edb6..c762c7747 100644
--- a/packages/cli/package.json
+++ b/packages/cli/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/cli",
-  "version": "0.13.47",
+  "version": "0.13.48",
   "description": "CLI for AgentsKit chat and project scaffolding.",
   "engines": {
     "node": ">=22"
diff --git a/packages/integrations/package.json b/packages/integrations/package.json
index 30fce8f52..f2333210b 100644
--- a/packages/integrations/package.json
+++ b/packages/integrations/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/integrations",
-  "version": "0.8.2",
+  "version": "0.8.3",
   "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 42ac0a0e3..c564b04d8 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@0.5.0 --tools fetch,search
+npx -y @agentskit/mcp@0.5.1 --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@0.5.0", "--tools", "fetch,search"] }
+    "agentskit": { "command": "npx", "args": ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"] }
   }
 }
 ```
@@ -52,7 +52,7 @@ Codex (`~/.codex/config.toml`):
 ```toml
 [mcp_servers.agentskit]
 command = "npx"
-args = ["-y", "@agentskit/mcp@0.5.0", "--tools", "fetch,search"]
+args = ["-y", "@agentskit/mcp@0.5.1", "--tools", "fetch,search"]
 ```
 
 OpenClaw:
@@ -61,7 +61,7 @@ OpenClaw:
 openclaw mcp add agentskit \
   --command npx \
   --arg -y \
-  --arg @agentskit/mcp@0.5.0 \
+  --arg @agentskit/mcp@0.5.1 \
   --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@0.5.0 --agents legal-contract-reviewer,fintech-kyc-screener --provider openai
+npx @agentskit/mcp@0.5.1 --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 106a854c8..9983602e9 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@0.5.0', '--tools', 'fetch,search'] as const
+const args = ['-y', '@agentskit/mcp@0.5.1', '--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@0.5.0 --tools fetch,search',
+    cli: 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search',
     config: {
       mcpServers: {
         agentskit: stdioServer,
@@ -30,7 +30,7 @@ export const codingAgentHostConfigs = {
     },
   },
   codex: {
-    cli: 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.0 --tools fetch,search',
+    cli: 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search',
     config: [
       '[mcp_servers.agentskit]',
       `command = "${command}"`,
@@ -63,7 +63,7 @@ export const codingAgentHostConfigs = {
   continue: {
     config: [
       'name: AgentsKit MCP',
-      'version: 0.5.0',
+      'version: 0.5.1',
       'schema: v1',
       'mcpServers:',
       '  - name: agentskit',
@@ -83,9 +83,9 @@ export const codingAgentHostConfigs = {
   },
 } as const
 
-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 expectedHostArgs = ['-y', '@agentskit/mcp@0.5.1', '--tools', 'fetch,search']
+const expectedClaudeCli = 'claude mcp add --scope project --transport stdio agentskit -- npx -y @agentskit/mcp@0.5.1 --tools fetch,search'
+const expectedCodexCli = 'codex mcp add agentskit -- npx -y @agentskit/mcp@0.5.1 --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),
-    `${host}.args must pin @agentskit/mcp@0.5.0 with fetch,search`,
+    `${host}.args must pin @agentskit/mcp@0.5.1 with fetch,search`,
   )
 }
 
@@ -195,7 +195,7 @@ const validateContinueYaml = (source: string): void => {
   const lines = source.split('\n')
   const metadata = [
     'name: AgentsKit MCP',
-    'version: 0.5.0',
+    'version: 0.5.1',
     '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 fa3f654d5..3c754fe68 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@0.5.0'
+const packageSpec = '@agentskit/mcp@0.5.1'
 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 e3a89ce07..959b9a7d6 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@0.5.0 --tools fetch,search
+npx -y @agentskit/mcp@0.5.1 --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 7c1487d0f..909a1c159 100644
--- a/packages/mcp/package.json
+++ b/packages/mcp/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/mcp",
-  "version": "0.5.0",
+  "version": "0.5.1",
   "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 2e9e88785..5f7623675 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@0.5.0', '--tools', 'fetch,search']
+const expectedArgs = ['-y', '@agentskit/mcp@0.5.1', '--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@0.5.0', '@agentskit/mcp')
+    codex.cli = String(codex.cli).replace('@agentskit/mcp@0.5.1', '@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 38e1743d2..0441ac046 100644
--- a/packages/memory/package.json
+++ b/packages/memory/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/memory",
-  "version": "0.12.0",
+  "version": "0.12.1",
   "description": "Persistent and vector memory backends for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/net/package.json b/packages/net/package.json
index 94c565e0c..8403dc7f7 100644
--- a/packages/net/package.json
+++ b/packages/net/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/net",
-  "version": "0.2.3",
+  "version": "0.2.4",
   "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 cc46b4dac..d6b42fb40 100644
--- a/packages/observability-langfuse/package.json
+++ b/packages/observability-langfuse/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@agentskit/observability-langfuse",
   "private": true,
-  "version": "0.2.27",
+  "version": "0.2.28",
   "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 5f30079f6..959cf82a5 100644
--- a/packages/observability/package.json
+++ b/packages/observability/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/observability",
-  "version": "0.13.0",
+  "version": "0.13.1",
   "description": "Logging and tracing layer for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/packages/rag/package.json b/packages/rag/package.json
index 6f1c5bc51..14e3052a2 100644
--- a/packages/rag/package.json
+++ b/packages/rag/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/rag",
-  "version": "0.7.0",
+  "version": "0.7.1",
   "description": "Plug-and-play retrieval-augmented generation for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/runtime/package.json b/packages/runtime/package.json
index 809f8e88e..3562d5322 100644
--- a/packages/runtime/package.json
+++ b/packages/runtime/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/runtime",
-  "version": "0.11.2",
+  "version": "0.11.3",
   "description": "Standalone agent runtime with ReAct loop for AgentsKit.",
   "keywords": [
     "agentskit",
diff --git a/packages/tools/package.json b/packages/tools/package.json
index 5701eb058..3c6577a9e 100644
--- a/packages/tools/package.json
+++ b/packages/tools/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@agentskit/tools",
-  "version": "0.13.15",
+  "version": "0.13.16",
   "description": "Reusable executable tools for AgentsKit agents.",
   "keywords": [
     "agentskit",
diff --git a/readme-standard-v1.json b/readme-standard-v1.json
index 67c26cd03..90cc590a6 100644
--- a/readme-standard-v1.json
+++ b/readme-standard-v1.json
@@ -285,7 +285,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:426fae28cef07a2c233f44dbdba94cdaf065b51058d50d1188662423b656a30d"
+        "sourceHash": "sha256:c1d12f117c66dcd88beab19771d4b66f4ed5cab3509aa8420375d2092c17c0a2"
       },
       "exceptions": []
     },
@@ -441,7 +441,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:588c7d93748863df07073908d1140d0b6ca0ca4a661d9a3e51e7ca2cc6d3adb7"
+        "sourceHash": "sha256:a9b76af30b86ad3000fdcd2d8afcd26bbfb03b9d0a95ec1783fb055da3190839"
       },
       "exceptions": []
     },
@@ -831,7 +831,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:fb5611839c65b08e8141e99bc1c1dbd200f8a5463d97c4eb9e66390dc9addc2a"
+        "sourceHash": "sha256:69f92ebdadc635fe4eaf2f83d17dff93aa0d141b192451d5c8d5a7d19c0f9e1f"
       },
       "exceptions": []
     },
@@ -886,7 +886,7 @@
       "commands": [
         {
           "id": "install",
-          "command": "npx -y @agentskit/mcp@0.5.0 --tools fetch,search",
+          "command": "npx -y @agentskit/mcp@0.5.1 --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:d5a1e09d6e1ef3833d2f7002a6f96037bc5552f4e1bf0d2009fcfdd82712fd3a"
+        "sourceHash": "sha256:efa08725014c871a28bf92af4a51911e1d96c894ee16b03b0ac29d0fa147c717"
       },
       "exceptions": []
     },
@@ -987,7 +987,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:bdf51bde4ff00a2e18b4d06c395e6ec7e261d3955e95fb4f47dfdb6c15ad9ffe"
+        "sourceHash": "sha256:3d4637166849499866ee56a33c672556620eb529659069504f74c2150aa1a0b5"
       },
       "exceptions": []
     },
@@ -1065,7 +1065,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:9f814d4ec52dee3148f5b3eb16314665e8634850f6acecaa9bbbb3a021f93a48"
+        "sourceHash": "sha256:7a759f1f696ec206d8571490da5a999559c432475f1e17a77fce22049b10adf1"
       },
       "exceptions": []
     },
@@ -1143,7 +1143,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:f19a8762bff4aba2147085ca69905ee084d2aaa0882add190110da6559ebe6c7"
+        "sourceHash": "sha256:fd6266953b482d6a0d57c23f3e0395cc131f5b2ec73f45b1ee19916e4e24c2b7"
       },
       "exceptions": []
     },
@@ -1221,7 +1221,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:c40fa0e67fe715897450f4411a5a759a231b6115277724ab2dc20945bf759ff8"
+        "sourceHash": "sha256:32896c9dc1acd00859265bc44a6d18bacdc4dc57ad4156d7246ca7f33a2c629b"
       },
       "exceptions": []
     },
@@ -1455,7 +1455,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:fb7092a785dc4e5dde4997eed722046958ba15debe76e15fa790ac19ddefa933"
+        "sourceHash": "sha256:725b70e60fb878378da8c67f17d9429942b55ca48e8aeeb100dc85c7dc1df70e"
       },
       "exceptions": []
     },
@@ -1689,7 +1689,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:416d932b79c8e868e04b3c497dacd166de5ec4025445b7a759fda1469761f333"
+        "sourceHash": "sha256:fe40a4da237cf1ec3ba43bef4b86548f443cb6efddd2749a6d9e6ae3f194ceee"
       },
       "exceptions": []
     },
@@ -2157,7 +2157,7 @@
           "docs/STABILITY.md",
           "scripts/readme-fixtures.test.mjs"
         ],
-        "sourceHash": "sha256:d6b8c271be5067ce5fb8d119016f9a39e8b50fa7e3f79dafc740b98f83c03db1"
+        "sourceHash": "sha256:d62305ec0cfecf597ba7accab06aed24b7f4b064794c4889faeab1004e777de9"
       },
       "exceptions": []
     },