diff --git a/CLAUDE.md b/CLAUDE.md index 55ccc14360..d703fa631b 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -1,5 +1,30 @@ # agentmemory — Agent Instructions +## Goal + +Put the right Memory in front of an Agent at the moment it needs it, and pay as little as +possible to be able to do that. Every feature here is judged against that sentence. + +- **Right content** — Recall returns what bears on the work in hand, not everything that + matched the query. A near-miss that costs the Agent a read is worse than one fewer result. +- **Right moment** — context arrives at hook boundaries (session start, pre-tool-use, + pre-compact) without the Agent having to know to ask. Memory the Agent must remember to + query is memory that goes unused. +- **Right scope** — results are bounded by project, branch, and Session + (`scope: "project" | "global"`, `src/types.ts:277`). One repo's work never surfaces in + another's. +- **Least record that restates** — store the smallest durable claim that reconstructs a + decision later: the conclusion and why, not the transcript that produced it. Observations + are raw and cheap; a Memory earns its place by being worth re-reading. +- **Durable beats recent** — a correction that changes future behaviour outranks a log of + what happened. Volume is a cost, not a measure of success; Eviction is expected, not a + failure mode. +- **Never on the critical path** — memory accelerates a Session and is never a dependency of + one. A dead daemon, a missed hook, or a slow Recall degrades the work; it does not block + it. +- **The Operator's attention is the scarce resource** — disk growth, restarts, and noisy + recall are the real costs, and they land on one person. + ## Architecture agentmemory is a persistent memory system for AI coding agents. It runs as a single process on one in-process Engine ([ADR 0001](./docs/adrs/0001-single-in-process-sqlite-engine.md)) — importing `src/index.ts` *is* starting the daemon. There is no separate runtime to install, spawn, adopt or stop. @@ -125,6 +150,11 @@ Hook scripts in `src/hooks/` are standalone Node.js scripts (no Engine import). - REST endpoints must whitelist fields — never pass raw request body to `sdk.trigger()` - Use `recordAudit()` for state-changing operations - Timestamps: capture once with `new Date().toISOString()` and reuse +- Outbound LLM/embedding calls go through `fetchWithTimeout`, which honors the configured + bound (`OPENAI_TIMEOUT_MS` → `AGENTMEMORY_LLM_TIMEOUT_MS` → 60s) exactly. The in-process + Engine has no invocation timeout, so never reintroduce a ceiling that clamps that bound — + the old 170s `HARD_BUDGET_CAP_MS` cited iii's 180s limit, silently cut every longer + timeout, and still reported the configured value in the error. ## Testing @@ -166,6 +196,23 @@ the `@agentmemory/mcp` shim wherever it is invoked as a proxy — `plugin/.mcp.j running server, not from the shim. The translated `READMEs/` were deleted rather than kept stale — do not re-add translations without a way to keep them current. +`docs/upstream-issue-triage.yaml` scores every open upstream issue by severity and fix +difficulty. It is a survey of defects in the shared code lineage, used to pick what is worth +fixing *here* — it is not a backlog to merge from, and it is a dated snapshot, not a live +mirror. Re-generate it from `gh issue list` rather than editing rows by hand. Its rows +cross-link the open upstream PRs that claim to fix them; `docs/upstream-pr-triage.yaml` +covers the remaining PRs — the ones that reference no open issue — scored by whether they +land in code this fork carries. Both are read for the defect and the diagnosis, never for +the patch. + +Every row in both files carries a `disposition`: `already-fixed` (verified against this tree; +`fixed_by`/`status` names the evidence), `wont-fix` (out of scope here — no iii engine, no npm +publish, no Windows CI, no deploy tree, or a host with no adapter in `src/cli/connect/`; +`wont_fix_reason` says which), or `candidate` — the working set. A row that was opened against +this tree also carries `status`, whose leading token says what the read found +(`fixed-here`, `present-here`, `partly-present-here`, `unresolved`). No `status` means +unchecked: the note is still upstream's claim, not a verified defect. + ## Current Stats (v0.9.29) - 55 MCP tools (all visible by default, `AGENTMEMORY_TOOLS=core` for the 8 essentials) diff --git a/CONTEXT.md b/CONTEXT.md index f95c59a16e..4e2d717a76 100644 --- a/CONTEXT.md +++ b/CONTEXT.md @@ -33,6 +33,12 @@ A durable distilled record derived from Observations, intended to be recalled la are content-addressed, so the same content captured twice is one Memory. _Avoid_: note, fact, item +**Session Summary**: +The distilled account of one Session, produced from its Observations at the Session's end. +A Session Summary is current until an Observation arrives after it was written, and a +current one is reused rather than reproduced. +_Avoid_: session digest, recap + **Entity**: A named thing the Agent worked with — a file, a symbol, a service, a concept — held as a node in the graph. diff --git a/docs/adrs/0001-single-in-process-sqlite-engine.md b/docs/adrs/0001-single-in-process-sqlite-engine.md index 0f9b2f73a7..2a8a435dad 100644 --- a/docs/adrs/0001-single-in-process-sqlite-engine.md +++ b/docs/adrs/0001-single-in-process-sqlite-engine.md @@ -66,3 +66,8 @@ this fork does not execute. Rejected on cost against zero realised benefit. registration and `trigger()` become direct in-process calls. - Documentation describing agentmemory as built on iii-engine's three primitives becomes false for this fork and must be rewritten, not annotated. +- Code that budgeted time against iii's 180s invocation timeout is now bounding itself + against a limit that no longer exists. `fetchWithTimeout`'s 170s `HARD_BUDGET_CAP_MS` was + one such residual: it clamped every configured LLM timeout above it while the error still + reported the unclamped value, so a slow local model looked like a provider failure. Treat + any remaining iii-derived ceiling as a defect, not a safety margin. diff --git a/docs/upstream-issue-triage.yaml b/docs/upstream-issue-triage.yaml new file mode 100644 index 0000000000..0603df299f --- /dev/null +++ b/docs/upstream-issue-triage.yaml @@ -0,0 +1,2734 @@ +# Upstream open-issue triage — rohitg00/agentmemory +# +# Scope: every open issue (not PR) on the upstream repository, scored on two axes. +# This fork does not merge from upstream (see CLAUDE.md); this is a survey of known +# defects in the shared code lineage, used to decide what is worth fixing here. +# +# severity — user impact if unfixed: +# high data loss, daemon/worker crash or unavailability, cross-agent or cross-user +# leakage, silent corruption, or a headline feature that never works at all +# med broken in a specific configuration, wrong or incomplete results, avoidable +# LLM spend, install/connect failure with a workaround +# low cosmetic, documentation, small UX, or a nice-to-have request +# +# difficulty — cost to fix in this codebase: +# low one or a few call sites, parameter plumbing, config or guard +# med several modules plus tests, a new endpoint, or a migration +# high architectural change, engine/dependency-level work, or a new subsystem +# +# kind — bug | feature | docs | chore | question | report | promo +# +# disposition — what this fork intends to do: +# already-fixed verified fixed in this tree; `fixed_by` names the evidence +# wont-fix out of scope here; `wont_fix_reason` says why +# candidate applies here and is unclaimed — the working set +# +# status — present only where the claim was checked against this tree by reading the +# code. "present-here" means the defect reproduces here as reported; +# "partly-present-here" means part of it was already fixed; "unresolved" means this +# tree cannot settle it. Absent means unchecked: the note is still upstream's claim. +# +# fix_prs — open upstream PRs whose title or body claims to close this issue +# (closing keyword). Read the PR before writing the fix here; several +# issues have competing PRs, and none of them are merged. +# mentioned_by_prs — open PRs that reference the issue without claiming to close it. +# Absent fields mean no open PR references the issue. 168 of the 291 open PRs reference +# no open issue at all and are NOT represented here. +# Snapshot: 308 open issues, cross-linked against 291 open PRs, as of +# 2026-09-21 (gh issue list / gh pr list --repo rohitg00/agentmemory). + +issues: + - number: 195 + title: "Architecture: Prevent Event Loop Freezes Caused by Heavy CPU-bound Tasks." + severity: high + difficulty: high + kind: bug + note: "Vector search blocks the single Node thread; fixing it means moving CPU work off the event loop." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/195 + - number: 259 + title: "Support custom metadata fields per session" + severity: low + difficulty: med + kind: feature + note: "Free-form session metadata plus a filterable surface in search_sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/259 + mentioned_by_prs: [306, 318] + - number: 262 + title: "Plugin doesn't work on OpenClaw 2026.5.7 — hook signatures and message_sent don't match" + severity: med + difficulty: med + kind: bug + note: "OpenClaw plugin hook signatures do not match the SDK, so chat capture is a no-op." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/262 + - number: 263 + title: "Consider implementing regex safety for Sentinels pattern matching" + severity: med + difficulty: low + kind: bug + note: "Sentinel patterns compile user regex directly; ReDoS on the observation path." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/263 + fix_prs: [297] + - number: 274 + title: "Lessons not created in flow-compress, consolidate, and consolidation-pipeline" + severity: med + difficulty: med + kind: bug + note: "flow-compress/consolidate extract lessons but never persist them via mem::lesson-save." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/274 + fix_prs: [279] + - number: 276 + title: "CRITICAL: Session summary and firstPrompt set to identical value" + severity: med + difficulty: low + kind: bug + note: "summary and firstPrompt written to the same truncated title (api.ts:542)." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/276 + - number: 299 + title: "Viewer UI shows empty state when fronted by a reverse proxy on standard ports (80/443) — JS hardcodes :3113" + severity: med + difficulty: low + kind: bug + note: "Viewer JS hardcodes :3113, so every panel is empty behind a reverse proxy." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/299 + mentioned_by_prs: [1083] + - number: 301 + title: "0.9.7 persistence fix is incomplete — distroless iii-engine (UID 65532) can't write to root-owned docker volume" + severity: high + difficulty: med + kind: bug + note: "Distroless UID 65532 cannot write the root-owned volume; state still non-persistent." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/301 + - number: 334 + title: "Antigravity (VS-code)" + severity: low + difficulty: med + kind: question + note: "Asks for an Antigravity plugin." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/334 + mentioned_by_prs: [1083] + - number: 336 + title: "Proposal: refactor viewer into a typed Vue/Vite frontend" + severity: low + difficulty: high + kind: feature + note: "Rewrite the single-file viewer as typed Vue 3 + Vite." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/336 + - number: 338 + title: "mem::graph-stats returns 'Function not found' — graph extraction broken in v0.9.11 / iii 0.11.2" + severity: high + difficulty: med + kind: bug + note: "mem::graph-stats returns Function not found; graph extraction dead in 0.9.11." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/338 + mentioned_by_prs: [341] + - number: 345 + title: "concept_edges table + BFS-depth-2 recall mode" + severity: med + difficulty: high + kind: feature + note: "concept_edges table plus a BFS-depth-2 recall mode." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/345 + fix_prs: [349] + - number: 392 + title: "Feature Request: Support time range filtering for memory_recall and memory_sessions" + severity: low + difficulty: med + kind: feature + note: "Time-range filters for memory_recall and memory_sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/392 + fix_prs: [414] + - number: 395 + title: "Bug: Local embedding stuck at \"none\" (BM25-only) in v0.9.13 despite correct xenova config" + severity: med + difficulty: med + kind: bug + note: "Local embedding stays at none despite a correct xenova config." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/395 + fix_prs: [412, 1283] + - number: 399 + title: "Feature request: Kubernetes Helm chart" + severity: low + difficulty: med + kind: feature + note: "Kubernetes Helm chart under deploy/helm." + disposition: wont-fix + wont_fix_reason: "Docker / PaaS / Kubernetes deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/issues/399 + - number: 400 + title: "Codex plugin install does not expose advertised full MCP tool surface" + severity: med + difficulty: low + kind: bug + note: "Codex plugin starts the shim without env, so only core tools are exposed." + disposition: wont-fix + wont_fix_reason: "unsupported host — Codex and Cursor are not supported here" + url: https://github.com/rohitg00/agentmemory/issues/400 + - number: 420 + title: "Feature: memory_signal_peek tool (read-only signal listing, no readAt mark)" + severity: low + difficulty: low + kind: feature + note: "Read-only memory_signal_peek that does not set readAt." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/420 + - number: 422 + title: "Viewer tab bar collapses/clips on Windows/Chromium layout" + severity: low + difficulty: low + kind: bug + note: "Viewer .tab-bar collapses vertically on Windows/Chromium." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/422 + - number: 429 + title: "How to move Agentmemory storage to another disk(e.g D or E)?" + severity: low + difficulty: low + kind: question + note: "How to relocate storage to another drive." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/429 + - number: 430 + title: "npm installs anthropic dependencies" + severity: low + difficulty: low + kind: bug + note: "Anthropic SDK pulled in for users who do not run Claude Code." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/430 + - number: 431 + title: "Feature Request: Native Hooks / Auto-Context Injection for OpenCode" + severity: med + difficulty: med + kind: feature + note: "Native hooks / auto context injection for OpenCode." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/431 + - number: 433 + title: "Expose agent_id / session_id in /agentmemory/audit so external rate-limiters can scope per-agent" + severity: med + difficulty: low + kind: feature + note: "Expose agent_id/session_id in /agentmemory/audit for per-agent rate limiting." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/433 + - number: 436 + title: "[Question] Desktop Apps Auto-detect?" + severity: low + difficulty: med + kind: question + note: "Whether desktop apps are auto-detected." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/436 + - number: 437 + title: "[BUG] Proxy Detected, Cannot Install" + severity: low + difficulty: low + kind: bug + note: "npm ECONNRESET behind a proxy on Windows 11; environment, not code." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/437 + - number: 439 + title: "support Volcengine embedding model URL" + severity: low + difficulty: low + kind: feature + note: "Support the Volcengine embedding base URL shape." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/439 + - number: 440 + title: "MCP `memory_recall` is aliased to smart-search and drops `format` param - full content unreachable via MCP" + severity: high + difficulty: low + kind: bug + note: "memory_recall aliased to smart-search and drops format, so full content is unreachable over MCP." + disposition: already-fixed + fixed_by: "memory_recall routes to mem::search, not smart-search (src/mcp/server.ts:145)." + url: https://github.com/rohitg00/agentmemory/issues/440 + - number: 455 + title: "Worker refuses to start after embedding provider change: stale index dimension mismatch (2048D vs 384D)" + severity: high + difficulty: med + kind: bug + note: "Stale index dimension mismatch makes the worker refuse to start; REST unreachable." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/455 + fix_prs: [575] + mentioned_by_prs: [525] + - number: 469 + title: "Vector Index Dimension Mismatch Crashes Server on Startup (2048 vs 384)" + severity: high + difficulty: med + kind: bug + note: "Same 2048-vs-384 dimension mismatch crashing startup on macOS." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/469 + fix_prs: [575] + mentioned_by_prs: [525] + - number: 480 + title: "Feature Request: Add eviction timer + env toggle for `mem::evict` (parity with `mem::auto-forget`)" + severity: med + difficulty: low + kind: feature + note: "Eviction timer plus env toggle for mem::evict, matching mem::auto-forget." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/480 + fix_prs: [490] + - number: 483 + title: "Feature request: i18n / Chinese localization for viewer UI" + severity: low + difficulty: high + kind: feature + note: "i18n for the viewer, starting with Chinese." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/483 + fix_prs: [541, 673] + - number: 496 + title: "Feature Request: Add DELETE endpoints for Actions and Crystals" + severity: med + difficulty: med + kind: feature + note: "DELETE endpoints for Actions and Crystals." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/496 + - number: 499 + title: "AgentMemory server becomes unresponsive with multiple concurrent Claude Code sessions" + severity: high + difficulty: high + kind: bug + note: "Server becomes unresponsive under three concurrent Claude Code sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/499 + - number: 505 + title: "Viewer Rebuild Graph button calls missing /agentmemory/graph/build endpoint" + severity: med + difficulty: low + kind: bug + note: "Viewer Rebuild Graph calls a nonexistent POST /agentmemory/graph/build." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/505 + fix_prs: [533, 538] + - number: 510 + title: "Bug: MCP tools not loaded by Claude Code despite healthy server and working plugin" + severity: med + difficulty: med + kind: bug + note: "MCP tools not loaded by Claude Code despite a healthy server." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/510 + - number: 511 + title: "npm install shows peer dependency conflict with @anthropic-ai/sdk (0.39.0 vs >=0.93.0)" + severity: low + difficulty: low + kind: bug + note: "Peer dependency conflict on @anthropic-ai/sdk during global install." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/511 + - number: 512 + title: "Chinese/Japanese i18n: improve viewer Memories search" + severity: med + difficulty: med + kind: feature + note: "CJK tokenization for viewer Memories search." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/512 + fix_prs: [536] + - number: 515 + title: "Codex worktrees are treated as separate projects, fragmenting context and project-scoped memory" + severity: med + difficulty: med + kind: bug + note: "Codex worktrees keyed by worktree cwd, fragmenting project-scoped memory." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/515 + fix_prs: [717] + - number: 518 + title: "AGENTMEMORY_SECRET enables api-auth middleware that also rejects internal worker callbacks, silently halting consolidation pipeline (v0.9.20)" + severity: high + difficulty: med + kind: bug + note: "AGENTMEMORY_SECRET auth rejects internal worker callbacks, silently halting consolidation." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/518 + fix_prs: [577] + mentioned_by_prs: [531, 535] + - number: 523 + title: "Suggestion: LAN Multi-Machine Support and Scheduled Database Backup" + severity: med + difficulty: high + kind: feature + note: "LAN multi-machine client/server mode plus scheduled database backup." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/523 + - number: 524 + title: "Feature request: built-in service install for daemon-mode autostart (macOS / Linux / Windows)" + severity: med + difficulty: med + kind: feature + note: "Ship LaunchAgent/systemd/Windows service units for daemon autostart." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/524 + - number: 527 + title: "Support OpenAI/Codex subscription-auth fallback without requiring OPENAI_API_KEY" + severity: low + difficulty: high + kind: feature + note: "OpenAI/Codex subscription-auth fallback without OPENAI_API_KEY." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/527 + - number: 552 + title: "docs: memory_lesson_save MCP tool exists but is completely undocumented (confidence, decay, reinforcement lifecycle)" + severity: low + difficulty: low + kind: docs + note: "memory_lesson_save exists but is absent from every doc surface." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/552 + - number: 557 + title: "error: unsupported OS when installing on Windows 11" + severity: med + difficulty: low + kind: bug + note: "iii console installer rejects MINGW64 on Windows 11." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/557 + - number: 559 + title: "agentmemory Graph empty" + severity: med + difficulty: med + kind: bug + note: "Graph stays empty on macOS with GRAPH_EXTRACTION_ENABLED=true." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/559 + - number: 586 + title: "Feature request: signal detector hook (on_turn_start) for entity extraction" + severity: low + difficulty: med + kind: feature + note: "Regex signal detector on on_turn_start for zero-LLM entity extraction." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/586 + - number: 589 + title: "How to use it with github copilot" + severity: low + difficulty: low + kind: question + note: "How to use agentmemory with GitHub Copilot in VS Code." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/589 + fix_prs: [622] + - number: 590 + title: "`mem::context` and `smart-search` skip `KV.insights` and `KV.semantic`" + severity: med + difficulty: med + kind: bug + note: "mem::context and smart-search never read KV.insights or KV.semantic." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/590 + - number: 595 + title: "Expose privacy-safe memory governance/delete receipts" + severity: low + difficulty: med + kind: feature + note: "Privacy-safe governance/delete receipts as an observability boundary." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/595 + - number: 596 + title: "OpenClaw plugin should scope recall/capture by session/user to avoid cross-user memory leakage" + severity: high + difficulty: med + kind: bug + note: "OpenClaw integration has no session/user scoping; cross-user memory leakage." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/596 + mentioned_by_prs: [598] + - number: 600 + title: "Bug: Knowledge graph extraction never fires automatically" + severity: high + difficulty: low + kind: bug + note: "event::session::stopped listens on a topic nothing publishes, so graph extraction never fires." + disposition: already-fixed + fixed_by: "Graph extraction fires from the observe pipeline (src/triggers/events.ts:123)." + url: https://github.com/rohitg00/agentmemory/issues/600 + - number: 603 + title: "cline --tui not recording sessions" + severity: med + difficulty: med + kind: bug + note: "cline --tui triggers no hooks, so no sessions are recorded." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/603 + - number: 605 + title: "memory_verify + Citation provenance: does a hybrid raw-anchor schema fit the audit lane?" + severity: low + difficulty: high + kind: feature + note: "Raw-anchor citation provenance schema for the verify/audit lane." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/605 + - number: 608 + title: "Doc - Link to an auto-generated architecture overview for agentmemory" + severity: low + difficulty: low + kind: promo + note: "Vendor offering to link auto-generated architecture docs." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/608 + - number: 609 + title: "Viewer dashboard can freeze Chrome by processing large mem-live sync backlogs" + severity: high + difficulty: med + kind: bug + note: "Large mem-live WebSocket sync backlog grows Chrome by GBs and freezes it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/609 + fix_prs: [623, 1302] + - number: 620 + title: "Reduce MCP tool surface area" + severity: med + difficulty: med + kind: feature + note: "Cut the MCP tool surface so models stop picking wrong entry points." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/620 + - number: 621 + title: "Standardize MCP tool dispatch test mocks" + severity: low + difficulty: low + kind: chore + note: "Standardize MCP dispatch test mocks on the repo pattern." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/621 + - number: 637 + title: "Windows: mem::summarize fails with Unicode characters (em dash) — ByteString conversion error" + severity: high + difficulty: low + kind: bug + note: "mem::summarize 500s on any character above U+00FF on Windows (ByteString conversion)." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/637 + - number: 645 + title: "安装时有一些警告,虽然不影响使用,但我觉得对新用户体验很差" + severity: low + difficulty: low + kind: bug + note: "Install-time npm peer dependency warnings; duplicate of 511." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/645 + - number: 655 + title: "memory_consolidate (semantic) and memory_reflect timeout due to sequential KV operations" + severity: high + difficulty: med + kind: bug + note: "Semantic consolidate and reflect time out from sequential KV operations." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/655 + mentioned_by_prs: [867] + - number: 658 + title: "Hermes saved plugin config does not affect runtime" + severity: med + difficulty: low + kind: bug + note: "Hermes saved plugin config is written but never read at runtime." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/658 + fix_prs: [663] + - number: 668 + title: "@agentmemory/mcp does not pin the matching core version, allowing fixed MCP behavior to regress" + severity: med + difficulty: low + kind: bug + note: "@agentmemory/mcp does not pin a matching core version, so fixed behavior regresses." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/668 + fix_prs: [1231] + - number: 691 + title: "[viewer] Graph nodes jumping/flickering on every poll refresh due to missing position preservation and no velocity cap" + severity: low + difficulty: med + kind: bug + note: "Graph nodes jump on every poll; no position preservation or velocity cap." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/691 + - number: 695 + title: "Feature request: Transparent fallback mode — indicate local storage in tool response & add strict/no-fallback option" + severity: med + difficulty: med + kind: feature + note: "Signal local-fallback writes in the tool response and add a strict no-fallback mode." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/695 + - number: 700 + title: "iii-engine state_store.db pollutes user's data/ directory — should use hidden path" + severity: med + difficulty: low + kind: bug + note: "Engine state_store.db lands in the user project's data/ directory." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/700 + - number: 706 + title: "Native audit: Zed auto-loads via Cline / AGENTS.md fallback" + severity: low + difficulty: med + kind: feature + note: "Use Zed's richer auto-load surface instead of MCP-only wiring." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/706 + - number: 708 + title: "Tracking: native-surface audit for the 5 agents added in PR #677 (skills/prompts/rules, not just MCP)" + severity: low + difficulty: med + kind: chore + note: "Tracking issue for native-surface audits of the five agents from PR 677." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/708 + - number: 712 + title: "iii-console install - Illegal option -o pipefail" + severity: med + difficulty: low + kind: bug + note: "iii-console installer uses set -o pipefail under dash on Ubuntu." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/712 + - number: 713 + title: "Bug: Local embedding fails with \"fetch failed\" despite model already downloaded" + severity: med + difficulty: med + kind: bug + note: "Local embedding fails with fetch failed although the model is already downloaded." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/713 + - number: 714 + title: "Any plans for OpenClaw plugin with hook-based auto-capture?" + severity: med + difficulty: med + kind: feature + note: "OpenClaw plugin with hook-based auto-capture, not just MCP tools." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/714 + - number: 715 + title: "Bug: `crypto.randomUUID()` breaks on Node.js < 19" + severity: low + difficulty: low + kind: bug + note: "crypto.randomUUID unavailable on Node 16; below the supported floor." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/715 + - number: 716 + title: "Resolve project identity from canonical repo URL, not directory basename" + severity: med + difficulty: med + kind: feature + note: "Resolve project identity from the canonical repo URL rather than the directory basename." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/716 + fix_prs: [717] + - number: 720 + title: "Token input cache rate dropped ~10pp due to per-turn system prompt mutation" + severity: med + difficulty: med + kind: bug + note: "Per-turn system prompt mutation drops the prompt cache hit rate ~10pp." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/720 + fix_prs: [1281] + mentioned_by_prs: [1325] + - number: 723 + title: "web端显示问题" + severity: low + difficulty: high + kind: feature + note: "Chinese localization request for the web viewer." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/723 + - number: 724 + title: "Reranker is a no-op: text-classification softmax over single-logit cross-encoder always returns 1.0" + severity: med + difficulty: med + kind: bug + note: "Reranker is a silent no-op: softmax over a single logit always scores 1.0." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/724 + - number: 725 + title: "local embedding provider hardcodes English-only model — make it configurable (multilingual default)" + severity: med + difficulty: low + kind: bug + note: "Local embedding provider hardcodes an English-only model, crippling non-English recall." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/725 + mentioned_by_prs: [943] + - number: 728 + title: "Hook resolveProject() is not worktree-aware: each git worktree becomes a separate memory silo" + severity: med + difficulty: low + kind: bug + note: "Hook resolveProject is not worktree-aware, so each worktree is its own memory silo." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/728 + - number: 730 + title: "feat: add Nix flake support for one-command installation" + severity: low + difficulty: med + kind: feature + note: "Nix flake for one-command installation." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/730 + fix_prs: [740] + - number: 732 + title: "feat(context): inject semantic + procedural + crystals + insights into mem::context (dark memory gap)" + severity: med + difficulty: med + kind: feature + note: "Inject semantic, procedural, crystals and insights into mem::context." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/732 + - number: 733 + title: "Project identity uses git-toplevel basename — collides across same-named repos, no remote-based identity" + severity: med + difficulty: med + kind: bug + note: "git-toplevel basename project identity collides across same-named repos." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/733 + fix_prs: [738] + - number: 736 + title: "feat(import): support Hermes Agent JSONL format in import-jsonl" + severity: low + difficulty: med + kind: feature + note: "Support the Hermes Agent JSONL schema in import-jsonl." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/736 + - number: 737 + title: "feat(recovery): abandoned session detection and backfill for abruptly terminated sessions" + severity: med + difficulty: med + kind: feature + note: "Detect and backfill sessions that terminated without a SessionEnd hook." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/737 + - number: 739 + title: "The memories endpoint `GET /agentmemory/memories` returns entries with an `id` field, so the data model already supports individual records." + severity: med + difficulty: low + kind: feature + note: "DELETE endpoint for individual memories by id." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/739 + - number: 745 + title: "`stop` hook prematurely calls `/agentmemory/session/end`, causing sessions to be marked as ended before user actually exits" + severity: med + difficulty: low + kind: bug + note: "stop hook ends the session on every assistant pause, marking live sessions completed." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/745 + fix_prs: [1288] + - number: 747 + title: "Same title consolidation produces duplicate and orphaned memories" + severity: med + difficulty: low + kind: bug + note: "Consolidation ignores memories created in the same run, producing duplicates and orphans." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/747 + - number: 751 + title: "memory_save accepts `project` parameter but iii engine serialization drops it — `memory-project-coverage` diagnose perpetually fails" + severity: high + difficulty: med + kind: bug + note: "Engine serialization drops Memory.project, so project coverage can never pass." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/751 + mentioned_by_prs: [1195] + - number: 754 + title: "Viewer: surface per-tier gate reasons when consolidation tabs are empty" + severity: low + difficulty: med + kind: feature + note: "Surface per-tier consolidation gate reasons instead of empty viewer tabs." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/754 + - number: 765 + title: "feat(consolidation): deterministic LLM extraction cache with prompt-SHA + model-snapshot keying" + severity: med + difficulty: high + kind: feature + note: "Deterministic LLM extraction cache keyed by prompt SHA and model snapshot." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/765 + - number: 766 + title: "feat(audit): consolidation/extract failures as first-class audit events with retry-with-backoff" + severity: med + difficulty: med + kind: feature + note: "Consolidation/extract failures as first-class audit events with backoff retry." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/766 + - number: 767 + title: "feat(budget): per-session token cap with soft warn + hard block" + severity: med + difficulty: med + kind: feature + note: "Per-session token cap with soft warn and hard block." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/767 + fix_prs: [934] + - number: 768 + title: "feat(graph): state trigger auto-stales targets on supersedes/contradicts/extends edges" + severity: low + difficulty: med + kind: feature + note: "State trigger that stales supersede/contradict targets automatically." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/768 + - number: 769 + title: "feat(bench): LongMemEval-S harness with statistical rigor + CI gate" + severity: med + difficulty: high + kind: feature + note: "LongMemEval-S retrieval benchmark harness with a CI gate." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/769 + - number: 770 + title: "feat(smart-search): index semantic facts, procedural skills, crystals, insights, facets" + severity: med + difficulty: med + kind: feature + note: "Index semantic, procedural, crystals, insights and facets in smart-search." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/770 + fix_prs: [784] + - number: 787 + title: "Project scoping not honored by MCP search tools (memory_smart_search / memory_recall) or /smart-search" + severity: high + difficulty: low + kind: bug + note: "MCP search tools and /smart-search ignore project, so reads are unscoped." + disposition: candidate + status: "present-here — mem::search accepts project (src/functions/search.ts:514) and mem::smart-search accepts it too (src/functions/smart-search.ts:88), but neither MCP handler forwards args.project — memory_recall sends only query/limit/format/token_budget/agentId (src/mcp/server.ts:145) and memory_smart_search only query/expandIds/limit (src/mcp/server.ts:346). Every MCP read is cross-project." + url: https://github.com/rohitg00/agentmemory/issues/787 + fix_prs: [869] + mentioned_by_prs: [1195] + - number: 790 + title: "Rebuild agent-memory.dev as a technical showcase" + severity: low + difficulty: high + kind: feature + note: "Rebuild agent-memory.dev as a technical showcase." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/790 + - number: 808 + title: "OpenRouter reasoning models need provider-specific reasoning request/response support" + severity: med + difficulty: med + kind: feature + note: "OpenRouter reasoning models need provider-specific request and response handling." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/808 + fix_prs: [810] + - number: 809 + title: "OpenRouterEmbeddingProvider hardcodes 1536 dimensions, breaking non-1536 embedding models" + severity: med + difficulty: low + kind: bug + note: "OpenRouterEmbeddingProvider hardcodes 1536 dimensions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/809 + fix_prs: [811, 856] + - number: 817 + title: "AGENT_ID isolated scope not enforced on mem::search (REST /search, memory_recall, recall_context) → cross-agent memory leak" + severity: high + difficulty: low + kind: bug + note: "mem::search applies no agent-scope filter, leaking memory across isolated agents." + disposition: already-fixed + fixed_by: "mem::search enforces agent scope; the code cites this issue (src/mcp/server.ts:137)." + url: https://github.com/rohitg00/agentmemory/issues/817 + fix_prs: [821] + mentioned_by_prs: [1201] + - number: 819 + title: "memory_smart_search returns results from all layers flat — no way to search only memory_save entries (L2)" + severity: med + difficulty: med + kind: feature + note: "No way to search only memory_save entries; observation noise buries L2." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/819 + - number: 820 + title: "memory_governance_delete returns deleted:0 for memory_save entries" + severity: med + difficulty: med + kind: bug + note: "memory_governance_delete returns deleted:0 for valid memory_save ids." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/820 + - number: 825 + title: "Bug: no path to rebuild or reset a >25K legacy graph — snapshot-rebuild + graph-reset both heartbeat-crash on state::list (follow-up to #816)" + severity: high + difficulty: high + kind: bug + note: "No path to rebuild or reset a >25K legacy graph; both paths heartbeat-crash." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/825 + - number: 826 + title: "feat: Expose post-tool-use and TaskCompleted hooks to Python plugin (Hermes integration)" + severity: med + difficulty: med + kind: feature + note: "Expose post-tool-use and TaskCompleted hooks to the Python/Hermes plugin." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/826 + - number: 828 + title: "Bug: entity-graph query/traversal (graph-query query + startNodeId) heartbeat-crashes past ~25K nodes on FRESH corpora — needs the per-node edge index #816 references" + severity: high + difficulty: high + kind: bug + note: "graph-query traversal enumerates the full graph and crashes past ~25K nodes." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/828 + fix_prs: [893] + - number: 833 + title: "/agentmemory:forget deletes nothing — memory_governance_delete only targets the memories store, not observations (reports success: true)" + severity: high + difficulty: med + kind: bug + note: "forget reports success while deleting nothing; governance delete never targets observations." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/833 + - number: 834 + title: "on_memory_write hook: smarter sync — skip add/update, archive on remove" + severity: low + difficulty: low + kind: feature + note: "Smarter on_memory_write sync: skip add/update, archive on remove." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/834 + fix_prs: [864] + - number: 843 + title: "All data lost on restart — stop order kills iii engine before worker can flush" + severity: high + difficulty: med + kind: bug + note: "Stop order kills the engine before the worker flushes; all data lost on restart." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/843 + mentioned_by_prs: [1268] + - number: 850 + title: "[Feature] Memory validation layer to detect poisoned/injected memories" + severity: med + difficulty: high + kind: feature + note: "Validation layer to detect poisoned or injected memories." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/850 + - number: 863 + title: "OPENAI_EMBEDDING_API_KEY" + severity: med + difficulty: low + kind: bug + note: "OPENAI_EMBEDDING_API_KEY ignored when OPENAI_API_KEY is also set." + disposition: already-fixed + fixed_by: "OPENAI_EMBEDDING_API_KEY takes precedence (src/providers/embedding/openai.ts:61)." + url: https://github.com/rohitg00/agentmemory/issues/863 + mentioned_by_prs: [943] + - number: 866 + title: "fix: MCP proxy CALL_TIMEOUT_MS hardcoded at 15 s causes memory_consolidate / memory_reflect to abort on large stores" + severity: med + difficulty: low + kind: bug + note: "MCP proxy CALL_TIMEOUT_MS hardcoded at 15s aborts consolidate and reflect." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/866 + fix_prs: [867] + - number: 871 + title: "Use ~/.local/bin per XDG Base Directory Specification instead of ~/.agentmemory/bin" + severity: low + difficulty: low + kind: chore + note: "Install iii to ~/.local/bin per XDG instead of ~/.agentmemory/bin." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/871 + - number: 873 + title: "Agent picker: selected `OpenCode` becomes `Claude Code`" + severity: med + difficulty: low + kind: bug + note: "Agent picker selection of OpenCode resolves to Claude Code." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/873 + - number: 874 + title: "doctor: \"iii-on-path-not-local-bin\" and fixes is a no-op -- install succeeds but check still finds other binary first" + severity: med + difficulty: low + kind: bug + note: "doctor fix for iii-on-path-not-local-bin never satisfies its own check." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/874 + fix_prs: [895] + - number: 875 + title: "engine-version-mismatch fix doesn't update PATH — iii-on-path-not-local-bin still fails after install" + severity: med + difficulty: low + kind: bug + note: "engine-version-mismatch fix does not update PATH, so doctor loops." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/875 + fix_prs: [895] + - number: 876 + title: "Lost all data after restart" + severity: high + difficulty: med + kind: bug + note: "All memories lost after a laptop restart; same class as 843." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/876 + - number: 880 + title: "Add Log Time" + severity: low + difficulty: low + kind: feature + note: "Show timestamps in the viewer log panel." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/880 + - number: 881 + title: "LocalEmbeddingProvider should respect EMBEDDING_MODEL env var for non-English embeddings" + severity: med + difficulty: low + kind: bug + note: "LocalEmbeddingProvider ignores EMBEDDING_MODEL, blocking multilingual embeddings." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/881 + fix_prs: [943] + - number: 884 + title: "MCP tool descriptions lack 'when to use' guidance — LLM struggles to choose the right tool among 53" + severity: med + difficulty: med + kind: feature + note: "Tool descriptions say what, not when, so the model picks wrong among 53." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/884 + fix_prs: [904] + - number: 888 + title: "MCP call path for slot tools returns opaque \"Internal error\" when AGENTMEMORY_SLOTS is not enabled (v0.9.27)" + severity: low + difficulty: low + kind: bug + note: "Slot tools return opaque Internal error when AGENTMEMORY_SLOTS is off." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/888 + fix_prs: [894] + - number: 889 + title: "memory_smart_search `expandIds` is a silent no-op: standalone MCP shim drops the param, and the engine cannot expand the `mem_…` IDs smart-search returns" + severity: med + difficulty: med + kind: bug + note: "expandIds is a no-op end to end: shim drops it and the engine cannot expand mem_ ids." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/889 + fix_prs: [1230] + - number: 890 + title: "mesh/export has no pagination — backlogs past the worker WS message limit fail forever with \"Invocation stopped\", making mesh sync unrecoverable after an outage" + severity: high + difficulty: med + kind: bug + note: "mesh/export has no pagination, so sync can never catch up after a long outage." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/890 + mentioned_by_prs: [1143] + - number: 899 + title: "feat: allow a cheaper model for compression than for summarization" + severity: med + difficulty: low + kind: feature + note: "Allow a cheaper model for compression than for summarization." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/899 + fix_prs: [900] + - number: 902 + title: "Compressed output shouldn't be locked to English: add language option to eliminate the semantic gap between query and memory" + severity: med + difficulty: low + kind: bug + note: "Compression prompt is hardcoded English, creating a query/memory semantic gap." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/902 + - number: 905 + title: "Example request: AIML API with OpenAI-compatible agent memory workflows" + severity: low + difficulty: low + kind: promo + note: "Vendor asking for an AIML API docs example." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/905 + - number: 906 + title: "Viewer stuck on \"Loading actions...\" when any action has tags as a string instead of an array" + severity: med + difficulty: low + kind: bug + note: "Viewer Actions tab hangs forever if any action has tags as a string." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/906 + - number: 907 + title: "Request to add available APIs for batch session deletion and graph cleanup" + severity: med + difficulty: med + kind: feature + note: "Batch session deletion and graph cleanup APIs." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/907 + - number: 908 + title: "Claude Code can't connect to agentmemory's mcp server anymore" + severity: med + difficulty: med + kind: bug + note: "Claude Code can no longer connect to the MCP server on 0.9.27." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/908 + - number: 909 + title: "agentmemory hangs ~90s on shutdown (systemd black screen) — iii-sdk telemetry shutdown never resolves" + severity: med + difficulty: med + kind: bug + note: "SIGTERM handler never resolves, hanging OS shutdown ~90s under systemd." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/909 + fix_prs: [910] + - number: 911 + title: "remember skill stores user secrets verbatim — W007 insecure credential handling" + severity: high + difficulty: med + kind: bug + note: "remember skill preserves exact phrasing, storing user secrets verbatim (Snyk W007)." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/911 + mentioned_by_prs: [941] + - number: 912 + title: "OpenCode not listed as supported agent in `connect` command" + severity: med + difficulty: low + kind: bug + note: "connect opencode fails with Unknown agent despite appearing in the picker." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/912 + - number: 913 + title: "Windows: engine state store silently falls back to in-memory (total data loss on restart) — relative `./data/` path + detached spawn without `cwd`" + severity: high + difficulty: med + kind: bug + note: "Windows scheduled-task CWD makes the engine run in-memory; total data loss per restart." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/913 + - number: 915 + title: "\"/recall\" slash command fails: HTTP 404 on memory_lesson_recall endpoint" + severity: med + difficulty: low + kind: bug + note: "/recall calls memory_lesson_recall, which has no server endpoint (404)." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/915 + - number: 918 + title: "GET /agentmemory/memories ignores the `project` query param — returns all projects" + severity: med + difficulty: low + kind: bug + note: "GET /agentmemory/memories ignores the project query param." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/918 + fix_prs: [988] + - number: 919 + title: "feat: add LLM_PROVIDER env var to explicitly select primary provider (bypass hardcoded detection order)" + severity: med + difficulty: low + kind: feature + note: "LLM_PROVIDER env var to bypass hardcoded provider detection order." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/919 + - number: 921 + title: "Viewer dashboard flickers under active sessions - per-event full reload blanks the panel" + severity: low + difficulty: med + kind: bug + note: "Per-event full reload blanks the dashboard, causing flicker under active sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/921 + - number: 924 + title: "[BUG] `import-jsonl` utility fails silently on historical data due to Claude Code cleanupPeriodDays conflict" + severity: low + difficulty: low + kind: docs + note: "import-jsonl finds nothing because Claude Code purges transcripts after cleanupPeriodDays." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/924 + - number: 925 + title: "graph search silently drops all results due to hardcoded `sessionId: \"\"`" + severity: high + difficulty: low + kind: bug + note: "Graph retrieval hardcodes sessionId:\"\", discarding every traversal result." + disposition: already-fixed + fixed_by: "GraphRetrieval.resolveSessionIds verifies each hit's owning session and falls back to a scan on a miss (src/functions/graph-retrieval.ts:72-126). No hardcoded sessionId; the only \"\" in the tree is src/replay/timeline.ts:107." + url: https://github.com/rohitg00/agentmemory/issues/925 + fix_prs: [937] + - number: 926 + title: "memory_save & memory_recall silently drop the `project` parameter (MCP server)" + severity: high + difficulty: low + kind: bug + note: "MCP memory_save and memory_recall silently drop project; memories saved unscoped forever." + disposition: already-fixed + fixed_by: "memory_save exposes project and agentId (src/mcp/tools-registry.ts:78-86)." + url: https://github.com/rohitg00/agentmemory/issues/926 + fix_prs: [1195] + - number: 931 + title: "Local embedding provider (@xenova/transformers) not detected in v0.9.27 — smart-search returns 0 semantic hits" + severity: med + difficulty: med + kind: bug + note: "Local embedding provider never activates in 0.9.27; smart-search silently BM25-only." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/931 + fix_prs: [1284] + mentioned_by_prs: [1283] + - number: 938 + title: "import \"replace\" wipes KV but leaves search indices stale" + severity: high + difficulty: med + kind: bug + note: "Replace-import wipes KV but leaves the search indices stale until restart." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/938 + - number: 944 + title: "[viewer] Agentmemory Token Savings always 0% — estInjected counts all sessions, not just active" + severity: low + difficulty: low + kind: bug + note: "Token Savings always 0% because estInjected counts all sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/944 + - number: 945 + title: "Missing lesson_delete tool — no way to delete/cleanup lessons via MCP or REST API" + severity: med + difficulty: med + kind: feature + note: "No lesson_delete tool or REST endpoint anywhere." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/945 + - number: 948 + title: "Add option to include low-score results in semantic search" + severity: low + difficulty: low + kind: feature + note: "Option to include low-score semantic results rather than returning nothing." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/948 + - number: 950 + title: "[BUG] Docker: iii-engine uses in_memory despite file_based config in /app/config.yaml" + severity: high + difficulty: med + kind: bug + note: "Docker engine uses in_memory despite a mounted file_based config." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/950 + - number: 951 + title: "Dashboard MEMORIES tile counts only memory_save facts, not the semantic/summary corpus (reads \"1\" on a healthy install)" + severity: low + difficulty: low + kind: bug + note: "Dashboard MEMORIES tile counts only memory_save facts, reading 1 on a healthy install." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/951 + fix_prs: [973] + - number: 952 + title: "agentmemory connect zed overwrites existing Zed settings.json instead of merging context_servers" + severity: high + difficulty: low + kind: bug + note: "connect zed overwrites the user's entire settings.json instead of merging." + disposition: already-fixed + fixed_by: "connect zed merges into context_servers (src/cli/connect/zed.ts:18)." + url: https://github.com/rohitg00/agentmemory/issues/952 + fix_prs: [966] + - number: 964 + title: "Local iii engine grows to ~3.7GB RSS with file_based storage and contributes to desktop OOMs" + severity: high + difficulty: high + kind: bug + note: "Engine grows to ~3.7GB RSS with file_based storage and contributes to desktop OOMs." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/964 + - number: 965 + title: "CEM Tree Memory: Branch-classified retrieval on BEAM (78%) and MABench (77.4% F1)" + severity: low + difficulty: high + kind: promo + note: "Vendor benchmark pitch for a tree-memory architecture." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/965 + - number: 969 + title: "import-jsonl (and live capture) hangs indefinitely on observations containing multibyte/CJK text" + severity: high + difficulty: med + kind: bug + note: "Observations containing CJK text hang import and live ingestion indefinitely." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/969 + - number: 972 + title: "Add an optional HVTracker trust badge to the README" + severity: low + difficulty: low + kind: promo + note: "Request to add a third-party trust badge to the README." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/972 + - number: 977 + title: "Feature Request: Publish Official Docker Images to Docker Hub" + severity: med + difficulty: med + kind: feature + note: "Publish official prebuilt Docker images." + disposition: wont-fix + wont_fix_reason: "Docker / PaaS / Kubernetes deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/issues/977 + fix_prs: [1354] + - number: 978 + title: "Batch/throttle graph extraction instead of triggering an LLM call on every Stop hook" + severity: med + difficulty: med + kind: bug + note: "Graph extraction fires an LLM call on every Stop hook rather than at real session end." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/978 + fix_prs: [1280] + - number: 979 + title: "Feature Request: Separate API configuration for LLM and embedding providers" + severity: med + difficulty: low + kind: feature + note: "Separate API key and base URL for LLM versus embedding providers." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/979 + - number: 981 + title: "Add first-class workspace namespace isolation" + severity: med + difficulty: high + kind: feature + note: "First-class workspace namespace isolation above project scope." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/981 + fix_prs: [982] + - number: 985 + title: "Feature Request: batch mode for background LLM routines via Anthropic Message Batches API" + severity: low + difficulty: high + kind: feature + note: "Use the Anthropic Message Batches API for background routines." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/985 + - number: 986 + title: "Feature Request: plain self-hosted Docker deploy template (no PaaS control plane)" + severity: low + difficulty: med + kind: feature + note: "Plain self-hosted Docker deploy template with no PaaS control plane." + disposition: wont-fix + wont_fix_reason: "Docker / PaaS / Kubernetes deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/issues/986 + fix_prs: [976] + - number: 987 + title: "fix: opencode plugin projectPath falls back to hash ID instead of filesystem path" + severity: med + difficulty: low + kind: bug + note: "opencode plugin falls back to a hash project id when ctx.worktree is null." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/987 + - number: 990 + title: "GET /agentmemory/memories?latest=true returns oldest memories first — viewer shows stale data when corpus > limit" + severity: med + difficulty: low + kind: bug + note: "memories?latest=true returns the oldest rows, so the viewer shows stale data." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/990 + - number: 991 + title: "SessionEnd hook reports \"Hook cancelled\" on exit (1.5s deferred exit outlives shutdown grace)" + severity: low + difficulty: low + kind: bug + note: "SessionEnd's 1.5s deferred exit outlives the shutdown grace, reporting Hook cancelled." + disposition: already-fixed + fixed_by: "Telemetry hooks use the documented fetch-and-unref'd-setTimeout exit pattern." + url: https://github.com/rohitg00/agentmemory/issues/991 + - number: 993 + title: "post-tool-use captures agentmemory MCP calls as observations, polluting recall results" + severity: med + difficulty: low + kind: bug + note: "post-tool-use records agentmemory's own MCP calls, polluting recall." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/993 + fix_prs: [994, 1044] + - number: 996 + title: "Zero-LLM mode: mem::summarize no-op skip (no_provider) inflates failureCount with false failures" + severity: low + difficulty: low + kind: bug + note: "Zero-LLM summarize no-op skips are recorded as failures, inflating failureCount." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/996 + - number: 998 + title: "Codex-visible hook skill should distinguish Codex hooks from Claude Code hooks" + severity: low + difficulty: low + kind: docs + note: "Hook skill wording misleads Codex users about which hooks exist." + disposition: wont-fix + wont_fix_reason: "unsupported host — Codex and Cursor are not supported here" + url: https://github.com/rohitg00/agentmemory/issues/998 + - number: 1003 + title: "Graph extraction always 404s: requests retired model `claude-3-haiku-20240307`, ignores ANTHROPIC_MODEL" + severity: high + difficulty: low + kind: bug + note: "Graph extraction requests the retired claude-3-haiku-20240307 and 404s, ignoring ANTHROPIC_MODEL." + disposition: already-fixed + fixed_by: "claude-3-haiku-20240307 appears nowhere in src/." + url: https://github.com/rohitg00/agentmemory/issues/1003 + - number: 1007 + title: "Feature request: auto-restore from latest snapshot on daemon startup" + severity: med + difficulty: med + kind: feature + note: "Auto-restore from the latest snapshot on daemon startup." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1007 + - number: 1008 + title: "POST /agentmemory/consolidate returns 500 with unhandled \"[object Object]\" error" + severity: med + difficulty: low + kind: bug + note: "POST /agentmemory/consolidate 500s with an unhandled [object Object] error." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1008 + fix_prs: [1041] + - number: 1009 + title: "Feature request: asymmetric embedding support (query_param / document_param for OpenAI-compatible endpoints)" + severity: med + difficulty: med + kind: feature + note: "Asymmetric query/document embedding parameters for OpenAI-compatible endpoints." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1009 + - number: 1013 + title: "All HTTP REST endpoints return 404 after iii-engine WebSocket reconnection (iii-engine #1796) — v0.19.2 only a partial fix, holds in v0.22.x" + severity: high + difficulty: high + kind: bug + note: "All REST endpoints 404 after an engine WebSocket reconnect; the process stays alive." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1013 + - number: 1015 + title: "Post-restart diagnostic: search can't find 646 backfill obs; vector/graph/rerank unverified" + severity: med + difficulty: high + kind: report + note: "Post-restart diagnostic: 646 backfill observations unfindable, vector/graph/rerank unverified." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1015 + - number: 1018 + title: "Feature request: Add memory_update Tool for Memory Versioning" + severity: med + difficulty: med + kind: feature + note: "memory_update tool for versioning instead of duplicate v1 memories." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1018 + fix_prs: [1038] + - number: 1020 + title: "Zero-LLM noop provider never matches the `mem::summarize` short-circuit because `ResilientProvider` renames it to `resilient(noop)` — summaries fail as `empty_provider_response` / `too_many_chunks_skipped` instead of a clean skip" + severity: med + difficulty: low + kind: bug + note: "ResilientProvider renames noop, so the summarize short-circuit can never fire." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1020 + fix_prs: [1040] + - number: 1022 + title: "GET /agentmemory/sessions ignores the limit query param, always returns full history" + severity: high + difficulty: low + kind: bug + note: "GET /agentmemory/sessions ignores limit and returned 5.38M characters into the caller's context." + disposition: candidate + status: "present-here — api::sessions reads no limit query param (src/triggers/api.ts:849-887): it lists every session, fetches every summary in batches of 10, and returns the lot. The standalone MCP client does send ?limit= (src/mcp/standalone.ts:230) — the server ignores it." + url: https://github.com/rohitg00/agentmemory/issues/1022 + mentioned_by_prs: [1034, 1352] + - number: 1024 + title: "Backfill corpus quality: 6c727be fix landed but LLM drain stalled, 33% LLM-compressed after 11h" + severity: med + difficulty: high + kind: report + note: "Backfill corpus quality fix landed but the LLM drain stalled at 33% coverage." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1024 + - number: 1025 + title: "[Cat B] Follow-up: 1867 obs stuck synthetic, drain throughput dropped 16x" + severity: med + difficulty: high + kind: report + note: "Follow-up: 1867 observations stuck synthetic, drain throughput down 16x." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1025 + - number: 1026 + title: "[Cat A] Defer: b61469b dead code + stale inline eval scripts" + severity: low + difficulty: low + kind: chore + note: "Dead code and stale inline eval scripts from a failed experiment." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1026 + mentioned_by_prs: [1083] + - number: 1027 + title: "[Cat C] Closed: corpus-quality goal achieved (R@10=100% on best queries)" + severity: low + difficulty: low + kind: report + note: "Informational: corpus-quality goal met, R@10 ~100% on best queries." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1027 + - number: 1029 + title: "2026-07-08: agentmemory restart left LLM drain stalled (6 unresolved problems)" + severity: med + difficulty: high + kind: report + note: "Six unresolved problems observed while restarting the daemon and resuming the drain." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1029 + - number: 1030 + title: "upgrade-backfill-compression.py progress.json drift: 207 obs marked done but missing compressionKind" + severity: low + difficulty: med + kind: bug + note: "Backfill progress.json says 1588 done but 207 have no compressionKind." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1030 + - number: 1031 + title: "\"agentmemory connect antigravity\" cannot detect Antigravity CLI (agy) v1.1.0" + severity: med + difficulty: low + kind: bug + note: "connect antigravity cannot detect the agy CLI v1.1.0 on PATH." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1031 + - number: 1035 + title: "pi: extension throws stale ctx after await getHealth() in refreshStatus" + severity: med + difficulty: low + kind: bug + note: "pi extension throws on stale ctx after awaiting getHealth in refreshStatus." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1035 + fix_prs: [1036] + - number: 1047 + title: "PostToolUse (and other) hooks crash with exit code 1 when payload JSON is null (unguarded data.xxx access)" + severity: med + difficulty: low + kind: bug + note: "Hooks crash with exit 1 when the payload JSON is literal null (unguarded data access)." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1047 + - number: 1048 + title: "feat: add support for gemini-embedding-2 and make Gemini embedding model configurable" + severity: low + difficulty: low + kind: feature + note: "Support gemini-embedding-2 and make the Gemini embedding model configurable." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1048 + - number: 1054 + title: "Agentmemory dashboard doesn't reflect renamed Codex sessions (Windows)" + severity: low + difficulty: med + kind: bug + note: "Dashboard does not reflect sessions renamed in Codex on Windows." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/1054 + - number: 1058 + title: "OpenClaw plugin: agent_end observe broken (missing project/cwd, raw-path project, zombie active sessions)" + severity: high + difficulty: med + kind: bug + note: "OpenClaw agent_end omits project/cwd, so turns are never persisted and sessions stay active." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1058 + mentioned_by_prs: [1252] + - number: 1059 + title: "backfill observation.concepts during graph-extract to unblock legacy consolidate under AUTO_COMPRESS=false" + severity: med + difficulty: med + kind: bug + note: "Synthetic compression hardcodes empty concepts, so legacy consolidate never groups." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1059 + - number: 1060 + title: "AGENTMEMORY_TOOLS=core in ~/.agentmemory/.env is ignored (getVisibleTools reads process.env directly)" + severity: med + difficulty: low + kind: bug + note: "getVisibleTools reads process.env directly, ignoring ~/.agentmemory/.env." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1060 + - number: 1062 + title: "Session stop triggers mem::summarize twice; unchanged sessions are fully re-summarized on every stop" + severity: high + difficulty: low + kind: bug + note: "Session stop dispatches mem::summarize twice, each re-summarizing from scratch." + disposition: candidate + status: "partly-present-here — The double dispatch is fixed: the Stop hook posts only /session/end and cites this class of bug (src/hooks/stop.ts:43). The re-summarize half is live — mem::summarize never reads KV.summaries before running (src/functions/summarize.ts:235-360), so every stop is a full LLM pass over every compressed observation." + url: https://github.com/rohitg00/agentmemory/issues/1062 + fix_prs: [1061] + - number: 1063 + title: "Session stop re-runs graph extraction over the full observation set even when nothing changed" + severity: med + difficulty: low + kind: bug + note: "Session stop re-extracts the full observation set with no change detection." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1063 + fix_prs: [1061, 1280] + - number: 1064 + title: "Phantom sessions: hooks + import-jsonl ingest Claude Code's machine-generated summarizer/Warmup transcripts as real sessions" + severity: med + difficulty: low + kind: bug + note: "Claude Code summarizer and Warmup transcripts ingested as real phantom sessions." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1064 + - number: 1065 + title: "lesson-recall default minConfidence (0.1) + decay + auto-prune make most auto-consolidated lessons invisible, with no config override" + severity: high + difficulty: low + kind: bug + note: "Default minConfidence plus decay hides 438 of 439 lessons from recall, with no override." + disposition: candidate + status: "present-here — mem::lesson-recall defaults minConfidence to 0.1 and drops anything below it (src/functions/lessons.ts:159,173). The decay sweep floors confidence at 0.05 (src/functions/lessons.ts:320), so a decayed lesson lands in 0.05-0.1 and is invisible at the default. Upstream's \"no override\" is wrong — the param is exposed." + url: https://github.com/rohitg00/agentmemory/issues/1065 + - number: 1066 + title: "memory_heal claims \"fixable\": true for project-scope backfill but the heal step does nothing ({\"fixed\":0,\"skipped\":0})" + severity: med + difficulty: med + kind: bug + note: "memory_heal reports project backfill fixable but fixes nothing." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1066 + - number: 1067 + title: "connect claude-code hardcodes ~/.claude.json / ~/.claude/settings.json, ignoring CLAUDE_CONFIG_DIR" + severity: med + difficulty: low + kind: bug + note: "connect claude-code hardcodes ~/.claude, ignoring CLAUDE_CONFIG_DIR; reports success but is inert." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1067 + - number: 1069 + title: "bug: session-end hook causes 'Hook cancelled' due to .unref() making setTimeout passive while fetch keeps event loop alive" + severity: low + difficulty: low + kind: bug + note: "unref'd timer does not force exit while fetch holds the loop; Hook cancelled on close." + disposition: already-fixed + fixed_by: "Same as #991." + url: https://github.com/rohitg00/agentmemory/issues/1069 + fix_prs: [1070] + - number: 1073 + title: "standalone.mjs: memory_save ignores title/tags, uses content as title — breaks keyword search" + severity: med + difficulty: low + kind: bug + note: "standalone memory_save drops title and tags and stores content as the title." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1073 + - number: 1077 + title: "import-jsonl in server-hosted mode is fundamentally confusing" + severity: low + difficulty: med + kind: docs + note: "import-jsonl path semantics are undocumented and confusing in server-hosted mode." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1077 + - number: 1080 + title: "Smart search cannot expand or consistently scope durable memories" + severity: med + difficulty: med + kind: bug + note: "expandIds only searches observations and project is inconsistently applied to durable memories." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1080 + fix_prs: [1091] + - number: 1081 + title: "Allow configurable external rerank endpoints" + severity: low + difficulty: med + kind: feature + note: "Configurable external Cohere-compatible rerank endpoint." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1081 + fix_prs: [1090] + - number: 1086 + title: "oh-my-pi (omp) support" + severity: low + difficulty: med + kind: feature + note: "Support for the oh-my-pi agent." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1086 + fix_prs: [1099] + - number: 1092 + title: "Memory dashboard truncates long content and entries cannot be opened" + severity: low + difficulty: low + kind: bug + note: "Dashboard truncates long memory content with no way to expand the entry." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1092 + - number: 1093 + title: "Idle CPU leak: `iii` worker threads busy-wait, growing to a full core over ~40h uptime" + severity: high + difficulty: high + kind: bug + note: "Worker threads busy-wait to a full core over ~40h uptime." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1093 + - number: 1094 + title: "`infer-memory-projects` permanently unfixable after a project directory move — path-keyed project identity marks every memory ambiguous" + severity: med + difficulty: med + kind: bug + note: "Path-keyed project identity makes infer-memory-projects permanently ambiguous after a directory move." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1094 + - number: 1097 + title: "Droid: MCP tools force approval at every autonomy level — no risk annotations" + severity: med + difficulty: low + kind: feature + note: "Droid MCP tools lack risk annotations, forcing approval at every autonomy level." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1097 + fix_prs: [1098] + - number: 1103 + title: "import-jsonl (and claude-bridge MEMORY.md path) also hardcode ~/.claude, ignoring CLAUDE_CONFIG_DIR" + severity: med + difficulty: low + kind: bug + note: "import-jsonl and the claude-bridge MEMORY.md path also ignore CLAUDE_CONFIG_DIR." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1103 + - number: 1104 + title: "Community Windows + Codex Desktop plugin (auto-start, MCP, hooks, self-healing troubleshooting skill)" + severity: med + difficulty: high + kind: feature + note: "Community Windows + Codex Desktop plugin with autostart, MCP, hooks and a troubleshooting skill." + disposition: wont-fix + wont_fix_reason: "unsupported host — Codex and Cursor are not supported here" + url: https://github.com/rohitg00/agentmemory/issues/1104 + - number: 1105 + title: "memory_action_create MCP tool drops createdBy, unlike lease/frontier tools which correctly pass agentId" + severity: low + difficulty: low + kind: bug + note: "memory_action_create never forwards createdBy, so every action is created by unknown." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1105 + fix_prs: [1107] + - number: 1106 + title: "AGENTMEMORY_TOOLS env var in ~/.agentmemory/.env is ignored — MCP server always exposes all tools" + severity: med + difficulty: low + kind: bug + note: "AGENTMEMORY_TOOLS in ~/.agentmemory/.env ignored; duplicate of 1060." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1106 + - number: 1108 + title: "\"project\"-scoped memory slots (project_context, pending_items, self_notes, session_patterns) are actually global, not per-project" + severity: high + difficulty: med + kind: bug + note: "project-scoped memory slots share one flat namespace, so they are global in practice." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1108 + fix_prs: [1337] + - number: 1114 + title: "Session summary chunk merge: `concepts` duplicated across chunks, and silently emitted empty when chunks fail" + severity: med + difficulty: low + kind: bug + note: "Summary concepts are concatenated without dedupe and silently emitted empty when chunks fail." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1114 + - number: 1115 + title: "Orphaned index generations accumulate when the previous manifest read fails — store grows until the server pegs a core while idle" + severity: high + difficulty: med + kind: bug + note: "Orphaned index generations accumulate when the manifest read fails; the store grows until a core pegs." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1115 + fix_prs: [1256, 1282] + - number: 1116 + title: "Viewer stores bearer token in sessionStorage — users must re-enter on every new tab" + severity: low + difficulty: low + kind: bug + note: "Viewer keeps the bearer token in sessionStorage, so it must be re-entered per tab." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1116 + - number: 1117 + title: "Viewer stores bearer token in sessionStorage — users must re-enter on every new tab" + severity: low + difficulty: low + kind: bug + note: "Duplicate of 1116." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1117 + - number: 1119 + title: "createEmbeddingProvider's \"openai\" case hardcodes OPENAI_API_KEY, defeating OPENAI_EMBEDDING_API_KEY fallback" + severity: med + difficulty: low + kind: bug + note: "Embedding factory passes OPENAI_API_KEY explicitly, defeating the OPENAI_EMBEDDING_API_KEY fallback." + disposition: already-fixed + fixed_by: "Same as #863." + url: https://github.com/rohitg00/agentmemory/issues/1119 + - number: 1124 + title: "mem::smart-search crashes the whole iii worker — one request unregisters all 271 functions for ~5s" + severity: high + difficulty: high + kind: bug + note: "One smart-search request unregisters all 271 functions for ~5s, taking the whole REST surface down." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1124 + - number: 1127 + title: "mem::graph-extract hangs to its own 180s invocation timeout on every batch, eventually stalling state::get and crashing the worker" + severity: high + difficulty: high + kind: bug + note: "graph-extract hangs to its 180s invocation timeout each batch, stalling state::get and crashing the worker." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1127 + fix_prs: [1128] + mentioned_by_prs: [1384] + - number: 1131 + title: "Session summarization is quadratic: the Stop hook re-summarizes the entire session every assistant turn" + severity: high + difficulty: med + kind: bug + note: "The Stop hook re-summarizes the whole session every assistant turn; cost is quadratic." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1131 + fix_prs: [1269] + - number: 1133 + title: "buildGraphClusters breaks out of the seed loop on an already-visited seed, collapsing insights into a few mega-clusters" + severity: med + difficulty: low + kind: bug + note: "buildGraphClusters breaks instead of continuing on a visited seed, collapsing insights into mega-clusters." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1133 + fix_prs: [1243, 1343] + - number: 1140 + title: "CLI `consolidate` run against a live daemon boots a second instance, then breaks the daemon REST routing (every /agentmemory/* route 404s)" + severity: high + difficulty: med + kind: bug + note: "CLI consolidate against a live daemon boots a second instance and 404s every REST route." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1140 + - number: 1142 + title: "GET /agentmemory/export drops the iii worker once the response passes 16 MiB, 404ing every endpoint" + severity: high + difficulty: med + kind: bug + note: "GET /agentmemory/export past 16 MiB drops the worker and 404s every endpoint." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1142 + mentioned_by_prs: [1143, 1233] + - number: 1148 + title: "Fix disabled slot MCP errors and Windows test portability" + severity: med + difficulty: med + kind: bug + note: "Disabled slot tools 500, diagnostics misclassify global memories, tests are not Windows-portable." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/1148 + fix_prs: [1149] + mentioned_by_prs: [1352] + - number: 1151 + title: "agentmemory stop in Docker mode signals the wrong PID — can kill Docker Desktop's backend entirely, and separately tears down unrelated compose services" + severity: high + difficulty: med + kind: bug + note: "agentmemory stop in Docker mode signals the wrong PID and can kill Docker Desktop's backend." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1151 + - number: 1153 + title: "iii-engine OOM-killed every ~61s: allocation spike scales with the container limit (0.11.2 and 0.11.5)" + severity: high + difficulty: high + kind: bug + note: "iii-engine OOM-killed on a ~61s cycle; the allocation spike scales with the container limit." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1153 + - number: 1157 + title: "Observation eviction leaves session counters and graph provenance stale" + severity: med + difficulty: med + kind: bug + note: "Eviction leaves session observationCount and graph provenance stale." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1157 + - number: 1158 + title: "Scheduled Git snapshots grow without a retention policy" + severity: med + difficulty: low + kind: feature + note: "Git snapshot repository has no age, count or size retention policy." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1158 + - number: 1159 + title: "REST shims silently drop agentId on /agentmemory/remember and /agentmemory/observe — per-request multi-agent scoping impossible over REST" + severity: high + difficulty: low + kind: bug + note: "REST shims drop agentId on remember and observe, making per-request multi-agent scoping impossible." + disposition: candidate + status: "partly-present-here — /agentmemory/remember does forward agentId (src/triggers/api.ts:1046-1048). /observe does not, but by design: mem::observe treats the session row as the source of truth and falls back to env AGENT_ID only when no row exists (src/functions/observe.ts:157-170)." + url: https://github.com/rohitg00/agentmemory/issues/1159 + fix_prs: [1202, 1272] + mentioned_by_prs: [1198, 1199, 1201] + - number: 1160 + title: "memoryToObservation() drops agentId — memories invisible to every agent-scoped search (residual hole in the #817 fix)" + severity: high + difficulty: low + kind: bug + note: "memoryToObservation drops agentId, hiding every memory from agent-scoped search." + disposition: already-fixed + fixed_by: "memoryToObservation carries agentId through, with a comment naming this exact bug (src/state/memory-utils.ts:85-88)." + url: https://github.com/rohitg00/agentmemory/issues/1160 + fix_prs: [1198] + mentioned_by_prs: [1187, 1272] + - number: 1161 + title: "integrations/openclaw: recall hook never fires on current OpenClaw (before_agent_start is not a typed plugin hook) + no agent identity threading" + severity: med + difficulty: med + kind: bug + note: "OpenClaw recall hook never fires: before_agent_start is not a typed plugin hook." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1161 + - number: 1162 + title: "Engine SDK port 49134 is inside the Linux ephemeral range — silent bind failure, infinite worker retry loop" + severity: high + difficulty: med + kind: bug + note: "Engine SDK port 49134 sits in the Linux ephemeral range; a stolen port means an infinite silent retry loop." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1162 + - number: 1163 + title: "import: observations persist but are never added to the BM25 index (invisible to smart-search)" + severity: high + difficulty: med + kind: bug + note: "Imported observations are never added to the BM25 index and have no reindex path." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1163 + mentioned_by_prs: [1187] + - number: 1165 + title: "Issues connecting to antigravity agy cli" + severity: med + difficulty: low + kind: bug + note: "connect cannot reach the antigravity agy CLI." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1165 + - number: 1166 + title: "memory_diagnose reports all checks PASS on a completely empty store — data loss is indistinguishable from a healthy store" + severity: med + difficulty: low + kind: bug + note: "memory_diagnose reports 15/15 PASS on an empty store, so data loss looks healthy." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1166 + - number: 1167 + title: "The \"backups\" feature never backs up memory — ~/.agentmemory/backups/ holds only connect config snapshots; no memory backup/restore command exists" + severity: med + difficulty: med + kind: bug + note: "~/.agentmemory/backups holds only connect config snapshots; no memory backup command exists." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1167 + - number: 1168 + title: "Unbounded sourceObservationIds / sourceMemoryIds are what push collections past the worker limit in #1142 / #1124" + severity: high + difficulty: med + kind: bug + note: "Unbounded provenance arrays are ~97% of collection bytes and are what push collections past the worker limit." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1168 + fix_prs: [1295, 1349] + mentioned_by_prs: [1233, 1353] + - number: 1169 + title: "agentmemory, analyzed by Claude and compared with other memory systems" + severity: low + difficulty: low + kind: promo + note: "Third-party analysis link and Discord invite." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1169 + - number: 1171 + title: "graph nodes: sourceObservationIds capped at creation but unbounded on merge" + severity: high + difficulty: low + kind: bug + note: "mergeNode re-unions sourceObservationIds with no cap, so provenance grows monotonically." + disposition: already-fixed + fixed_by: "mergeNode runs the union through capSourceIds (src/functions/graph.ts:432); src/functions/graph-provenance.ts caps from the tail so merges keep the newest ids." + url: https://github.com/rohitg00/agentmemory/issues/1171 + fix_prs: [1294, 1349] + mentioned_by_prs: [1353] + - number: 1173 + title: "prompt_submit observations silently dropped: dedup hash ignores prompt content (all prompts share one hash)" + severity: high + difficulty: low + kind: bug + note: "prompt_submit dedup hash reads a nonexistent field, so all prompts share a hash and only the first in 300s survives." + disposition: already-fixed + fixed_by: "The dedup hash falls back to the whole payload when tool_input is absent (src/functions/observe.ts:76-81), so prompt_submit events hash on their own prompt." + url: https://github.com/rohitg00/agentmemory/issues/1173 + - number: 1178 + title: "Test suite reads the developer's real ~/.agentmemory/.env — 14 tests fail on any machine with an install" + severity: med + difficulty: low + kind: bug + note: "Tests read the developer's real ~/.agentmemory/.env; 14 fail on any machine with an install." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1178 + - number: 1179 + title: "No way to rebuild the search index without restarting: rebuildIndex only runs when the index is empty" + severity: med + difficulty: med + kind: feature + note: "rebuildIndex only runs at boot on an empty index; no supported rebuild path." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1179 + - number: 1180 + title: "Index persistence is never scheduled after boot: everything indexed at runtime is lost unless the process exits cleanly" + severity: high + difficulty: med + kind: bug + note: "Nothing schedules an index save after boot, so runtime indexing is lost unless the process exits cleanly." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1180 + - number: 1181 + title: "mem:audit is 71% index_persist rows the audit policy does not cover" + severity: med + difficulty: low + kind: bug + note: "index_persist rows are 71% of the audit store, slowing memory_audit and every startup read." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1181 + fix_prs: [1182] + - number: 1183 + title: "[FEAT][08-12-2026] add support on these `features` that I've notice so far using `agentmemory`, if not integrated yet." + severity: med + difficulty: high + kind: feature + note: "Grab bag: Windows connect support, Antigravity variants, doctor checks." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1183 + - number: 1184 + title: "opencode plugin: auto-title generation consumes the once-per-session memory injection — main conversation never gets context" + severity: high + difficulty: low + kind: bug + note: "OpenCode auto-title generation consumes the once-per-session injection, so the real conversation gets no context." + disposition: candidate + status: "unresolved — contextInjectedSessions is keyed by session id and set on the first experimental.chat.system.transform call (plugin/opencode/agentmemory-capture.ts:665-684). No guard distinguishes a title-generation pass from the real conversation. Whether the title pass shares the session id is an OpenCode runtime question this tree cannot answer." + url: https://github.com/rohitg00/agentmemory/issues/1184 + fix_prs: [1281] + - number: 1188 + title: "OpenCode sessions inherit project/cwd from shared module state across directories" + severity: high + difficulty: low + kind: bug + note: "OpenCode plugin keeps project and cwd in module state, recording sessions under another repository." + disposition: already-fixed + fixed_by: "Attribution is per-session via sessionProjects, resolved from each session's own directory; module-level defaults are only a fallback, and the comment names this bug (plugin/opencode/agentmemory-capture.ts:66-81)." + url: https://github.com/rohitg00/agentmemory/issues/1188 + fix_prs: [1281] + - number: 1189 + title: "Bug: graph-reset leaves pre-reset rows queryable and races graph-extract" + severity: med + difficulty: med + kind: bug + note: "graph-reset leaves pre-reset rows queryable and races a concurrent graph-extract." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1189 + - number: 1190 + title: "Bug: snapshot restore is additive and omits durable stores" + severity: high + difficulty: med + kind: bug + note: "Snapshot restore is additive and omits durable stores, so it is not a recovery point." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1190 + - number: 1191 + title: "Semantic and Procedural Memory need a bounded retention path" + severity: med + difficulty: med + kind: feature + note: "Semantic and procedural memory have no bounded retention path." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1191 + - number: 1192 + title: "Audit storage needs retention and bounded queries" + severity: med + difficulty: med + kind: feature + note: "Audit storage needs retention and bounded queries." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1192 + - number: 1193 + title: "Lesson and Insight tombstones need bounded physical reclamation" + severity: med + difficulty: med + kind: feature + note: "Lesson and insight tombstones need physical reclamation." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1193 + - number: 1196 + title: "Replay loses raw prompt/tool/response content under default synthetic compression" + severity: med + difficulty: med + kind: bug + note: "Default synthetic compression caps the narrative at 400 chars, so replay loses raw content." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1196 + - number: 1197 + title: "memory_save MCP path drops agentId at three more points — schema never exposes it, and two forwarding hops drop it independently of #1159/#1160" + severity: high + difficulty: low + kind: bug + note: "memory_save drops agentId at three more MCP-path points beyond 1159 and 1160." + disposition: already-fixed + fixed_by: "Same as #926." + url: https://github.com/rohitg00/agentmemory/issues/1197 + mentioned_by_prs: [1198, 1199, 1200, 1201, 1202, 1272] + - number: 1203 + title: "mem::summarize is dispatched twice per Stop hook" + severity: high + difficulty: low + kind: bug + note: "Two concurrent mem::summarize runs per Stop hook with no lease; duplicate of 1062." + disposition: already-fixed + fixed_by: "The Stop hook makes one call — /session/end — and session/end fans the summary out server-side (src/hooks/stop.ts:43, src/triggers/events.ts:99)." + url: https://github.com/rohitg00/agentmemory/issues/1203 + mentioned_by_prs: [1269] + - number: 1204 + title: "logger writes info and warn to stderr, so log platforms tag every line as an error" + severity: low + difficulty: low + kind: bug + note: "logger sends info and warn to stderr, so log platforms tag every line an error." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1204 + - number: 1208 + title: "feat: DeepSeek Harness (dsh) integration — connect adapter + cordis plugin" + severity: low + difficulty: med + kind: feature + note: "DeepSeek Harness connect adapter plus a cordis capture plugin." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1208 + fix_prs: [1209] + - number: 1218 + title: "Cannot update" + severity: med + difficulty: low + kind: bug + note: "npx upgrade does not move to the latest version." + disposition: wont-fix + wont_fix_reason: "npm distribution — nothing is published from this fork" + url: https://github.com/rohitg00/agentmemory/issues/1218 + - number: 1219 + title: "Bug: v0.9.29 default OpenAI model (`gpt-5.6-luna`) is a reasoning model, but `OpenAIProvider` sends `max_tokens` → every `mem::compress` / `mem::summarize` 400s" + severity: high + difficulty: low + kind: bug + note: "0.9.29's default OpenAI model is a reasoning model but the provider still sends max_tokens; every compress and summarize 400s." + disposition: candidate + status: "present-here — Default model is gpt-5.6-luna (src/providers/openai.ts:12, src/config.ts:95) and the chat body always sends max_tokens (src/providers/openai.ts:82) while also sending reasoning_effort (src/providers/openai.ts:95) — the two are mutually exclusive on OpenAI's chat-completions API. Nothing maps to max_completion_tokens." + url: https://github.com/rohitg00/agentmemory/issues/1219 + fix_prs: [1275] + - number: 1221 + title: "Graph node identity is the absolute path, so git worktrees and agent scratch dirs create duplicate nodes for the same file" + severity: med + difficulty: med + kind: bug + note: "Graph file-node identity is the absolute path, duplicating nodes across worktrees." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1221 + - number: 1223 + title: "Health reports critical from heapUsed/heapTotal, so /agentmemory/health returns 503 on a healthy process" + severity: med + difficulty: low + kind: bug + note: "Health grades memory against heapTotal, so a healthy process returns 503." + disposition: already-fixed + fixed_by: "Saturation is measured against the heap ceiling (src/health/thresholds.ts:88)." + url: https://github.com/rohitg00/agentmemory/issues/1223 + fix_prs: [1285] + mentioned_by_prs: [1236, 1237] + - number: 1225 + title: "MCP tools ask the model for a project id it has no way to know" + severity: high + difficulty: med + kind: bug + note: "Every project-scoped MCP tool asks the model for a project id nothing tells it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1225 + - number: 1226 + title: "mem::patterns walks every observation in the store — memory_patterns 500s \"Invocation stopped\" and drops the worker" + severity: high + difficulty: med + kind: bug + note: "mem::patterns loads every observation, 500s, and takes the worker with it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1226 + - number: 1228 + title: "Unable to recompress raw observations; compression failures leave observations uncompressed with no recovery path" + severity: med + difficulty: med + kind: feature + note: "No way to recompress an observation whose LLM compression failed." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1228 + - number: 1229 + title: "Viewer renders session summary as '[object Object]' (array fields not serialized)" + severity: low + difficulty: low + kind: bug + note: "Viewer renders array-valued session summaries as [object Object]." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1229 + fix_prs: [1332] + - number: 1232 + title: "memory_recall: token_budget silently drops the highest-value records first, making explicit memories look absent" + severity: high + difficulty: low + kind: bug + note: "token_budget fills with short high-scoring observations, excluding saved memories entirely." + disposition: candidate + status: "present-here — applyTokenBudget returns at the first item that overflows instead of skipping it (src/functions/search.ts:756-773), so one oversized record truncates everything ranked below it." + url: https://github.com/rohitg00/agentmemory/issues/1232 + fix_prs: [1242] + - number: 1235 + title: "Health reports cpu_critical on multi-core hosts: CPU percent is single-core scale but thresholds assume total capacity" + severity: med + difficulty: low + kind: bug + note: "CPU percent is single-core scale but thresholds assume total capacity, so multi-core hosts report critical." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1235 + fix_prs: [1236, 1286] + mentioned_by_prs: [1237] + - number: 1238 + title: "`GRAPH_EXTRACTION_ENABLED=false` does not stop graph writes - structural extraction runs unconditionally on session stop" + severity: med + difficulty: low + kind: bug + note: "GRAPH_EXTRACTION_ENABLED=false still runs structural extraction and writes graph rows." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1238 + - number: 1239 + title: "`POST /agentmemory/graph/reset` reports success while leaving nodes and edges untouched" + severity: high + difficulty: low + kind: bug + note: "POST /agentmemory/graph/reset reports success while leaving every node and edge in place." + disposition: already-fixed + fixed_by: "mem::graph-reset stamps resetAt on an empty snapshot, clears the name shards and flips retrieval onto the resetAt-filtering index path (src/functions/graph.ts:1308-1335); persistGraphDelta drops pre-reset rows from the merge (src/functions/graph.ts:784-791)." + url: https://github.com/rohitg00/agentmemory/issues/1239 + - number: 1240 + title: "`mem::summarize` asks for freehand XML and never sends `response_format` - 19% first-attempt parse failure, plus an unreported `narrative: Too small` validation failure" + severity: med + difficulty: med + kind: bug + note: "summarize asks for freehand XML with no response_format; 19% first-attempt parse failure." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1240 + - number: 1244 + title: "Session summarisation re-runs on already-summarised sessions (954x on one session; 150x more calls than material)" + severity: high + difficulty: med + kind: bug + note: "Summarisation re-runs on already-summarised sessions, 954x on one session, each a paid call." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1244 + - number: 1245 + title: "--instance / --port never reach the generated engine config, so a second daemon always binds instance-0 ports" + severity: high + difficulty: med + kind: bug + note: "--instance and --port never reach the generated engine config, so a second daemon always collides." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1245 + - number: 1249 + title: "vision-search queryText always fails: CLIP text branch uses feature-extraction, which demands pixel_values" + severity: med + difficulty: med + kind: bug + note: "vision-search queryText always fails: the CLIP text branch demands pixel_values." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1249 + fix_prs: [1250] + mentioned_by_prs: [1253] + - number: 1251 + title: "session/end creates malformed row when session does not exist" + severity: high + difficulty: low + kind: bug + note: "session/end upserts a malformed row for unknown sessions, later breaking GET /sessions." + disposition: candidate + status: "present-here — api::session::end calls kv.update on the session id with no existence check (src/triggers/api.ts:667-670), and update on a missing key auto-creates it as {} (src/engine/inproc/state.ts:11). An unknown id yields a row with only endedAt and status — which api::sessions then returns." + url: https://github.com/rohitg00/agentmemory/issues/1251 + fix_prs: [1252] + - number: 1254 + title: "vision-search silently ignores `limit` while every other search function uses it" + severity: low + difficulty: low + kind: bug + note: "vision-search silently ignores limit while every sibling honors it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1254 + fix_prs: [1255] + - number: 1264 + title: "Test suite: 30 failures on Windows, all test-side platform assumptions" + severity: med + difficulty: med + kind: bug + note: "30 Windows test failures, all test-side platform assumptions." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/1264 + mentioned_by_prs: [1265, 1266, 1267] + - number: 1270 + title: "Hooks with no tool payload produce empty compression prompts, burning LLM calls on content-free observations" + severity: med + difficulty: low + kind: bug + note: "Payload-free hooks build empty compression prompts, burning LLM calls on nothing." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1270 + - number: 1271 + title: "`getXmlTag` is a first-match unanchored regex, so any text before the payload can be parsed as the payload -- and `compressWithRetry` commits a response its own validator rejected" + severity: high + difficulty: med + kind: bug + note: "Unanchored getXmlTag takes the first match anywhere and compressWithRetry commits a response its validator rejected." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1271 + - number: 1273 + title: "Governance delete finds a captured observation but deletes nothing (deleted:0)" + severity: high + difficulty: med + kind: bug + note: "Governance delete finds a captured observation and deletes nothing." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1273 + - number: 1274 + title: "Feature: deprecate / supersede observations (invalidate bad captures) with audit trail" + severity: med + difficulty: med + kind: feature + note: "No way to deprecate or supersede a bad capture with an audit trail." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1274 + - number: 1276 + title: "Bug: Azure content filter rejections open the circuit breaker, failing unrelated compressions" + severity: med + difficulty: low + kind: bug + note: "Content-filter rejections count as provider failures and open the circuit breaker for unrelated work." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1276 + fix_prs: [1277] + - number: 1278 + title: "PreToolUse context injection cannot reach the model: raw stdout instead of the hook envelope, plus a 2s fetch budget against a 6-29s /agentmemory/enrich" + severity: high + difficulty: med + kind: bug + note: "PreToolUse injection uses raw stdout instead of the hook envelope, against a 2s budget on a 6-29s endpoint." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1278 + - number: 1279 + title: "[viewer] Dashboard \"Memories\" stat card caps at 500 — uses .length of a limited fetch instead of the total field" + severity: low + difficulty: low + kind: bug + note: "Dashboard Memories card uses the length of a limited fetch, capping at 500." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1279 + fix_prs: [1347] + - number: 1292 + title: "deriveCrystalAndLessons slices lessons mid-sentence: 775/981 (79%) unusable, and they cannot decay out" + severity: high + difficulty: med + kind: bug + note: "deriveCrystalAndLessons slices lessons mid-sentence; 775 of 981 unusable and some invert meaning." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1292 + - number: 1297 + title: "Feature request: signal deletion / TTL for signals" + severity: low + difficulty: med + kind: feature + note: "Signals have no delete or TTL, so the store grows without bound." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1297 + - number: 1312 + title: "[deps] Please bump iii-sdk / iii-engine past 0.11.2 — unbounded RSS growth + hung graceful shutdown in long-running deployments" + severity: med + difficulty: high + kind: chore + note: "Bump iii-sdk/engine past the 0.11.2 pin for RSS growth and shutdown hangs." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1312 + - number: 1313 + title: "benchmark/COMPARISON.md documents four npm scripts that are not defined" + severity: low + difficulty: low + kind: docs + note: "benchmark/COMPARISON.md documents four npm scripts that do not exist." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1313 + - number: 1314 + title: "COMPARISON.md states all datasets are committed; benchmark/data/.gitignore excludes the LongMemEval-S dataset" + severity: low + difficulty: low + kind: docs + note: "COMPARISON.md claims all datasets are committed; the LongMemEval-S dataset is gitignored." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1314 + - number: 1315 + title: "Question: where is the DCO sign-off requirement enforced?" + severity: low + difficulty: low + kind: question + note: "Where the CONTRIBUTING DCO sign-off requirement is actually enforced." + disposition: wont-fix + wont_fix_reason: "upstream housekeeping, marketing or repo diary" + url: https://github.com/rohitg00/agentmemory/issues/1315 + - number: 1316 + title: "Feature request: target-scoped graph provenance correction after agent mistakes" + severity: med + difficulty: high + kind: feature + note: "Target-scoped graph provenance correction without a full graph reset." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1316 + - number: 1321 + title: "Codex prompt_submit observation can be returned by same-turn recall (self-recall)" + severity: med + difficulty: low + kind: bug + note: "Same-turn recall can return the prompt the hook just persisted." + disposition: wont-fix + wont_fix_reason: "unsupported host — Codex and Cursor are not supported here" + url: https://github.com/rohitg00/agentmemory/issues/1321 + - number: 1326 + title: "/sessions endpoint still hangs indefinitely in v0.9.29 (fix #1136 installed) — kv.list(KV.sessions) never resolves, breaks viewer dashboard + MCP memory_sessions" + severity: high + difficulty: high + kind: bug + note: "GET /agentmemory/sessions still hangs indefinitely on 0.9.29; kv.list(KV.sessions) never resolves." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1326 + - number: 1328 + title: "feat: Allow custom HTTP headers on LLM provider requests (needed for x-opencode-session)" + severity: med + difficulty: low + kind: feature + note: "Custom HTTP headers on LLM provider requests, needed for x-opencode-session." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1328 + - number: 1330 + title: "Docker deployment: real-time viewer on :3113 is unreachable (engine-only image never launches the npm viewer process)" + severity: med + difficulty: med + kind: bug + note: "The Docker engine-only image never launches the viewer, so :3113 is unreachable." + disposition: wont-fix + wont_fix_reason: "Docker / PaaS / Kubernetes deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/issues/1330 + - number: 1331 + title: "Windows: ~/.agentmemory/.env never loaded (HOME unset), so AGENTMEMORY_SECRET is empty and every integration fails silently while reporting healthy" + severity: high + difficulty: med + kind: bug + note: "Windows never loads ~/.agentmemory/.env, so every authenticated request 401s while reporting healthy." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/1331 + - number: 1333 + title: "Pinned slot block is silently dropped whole when it exceeds TOKEN_BUDGET (continue, not truncate)" + severity: med + difficulty: low + kind: bug + note: "The pinned-slot block is dropped whole and silently when it exceeds the token budget." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1333 + - number: 1334 + title: "/agentmemory/export returns the oversized refusal under HTTP 200, and materialises the payload before refusing it (18.5 GB RSS)" + severity: high + difficulty: med + kind: bug + note: "The oversized export refusal ships under HTTP 200 after materialising 18.5GB of RSS." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1334 + - number: 1335 + title: "Shutdown index save is lost to a container teardown race, and boot only reconciles a completely empty index" + severity: high + difficulty: med + kind: bug + note: "The shutdown index save is lost to a teardown race and boot only reconciles a fully empty index." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1335 + - number: 1338 + title: "fix(providers): createFallbackProvider drops baseURL — custom-endpoint fallbacks (DeepSeek/Kimi/etc.) hit the wrong API host" + severity: med + difficulty: low + kind: bug + note: "createFallbackProvider drops baseURL, so custom-endpoint fallbacks hit the vendor's host." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1338 + fix_prs: [1345] + - number: 1339 + title: "graph/build: server-side batch loop outlives its \"Invocation stopped\" HTTP response; bulk backfill impossible on large graph stores (Win32 + remote LLM, corroborates #1127)" + severity: high + difficulty: high + kind: bug + note: "graph/build's server-side batch loop outlives its response and burns LLM calls for hours." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1339 + mentioned_by_prs: [1384] + - number: 1340 + title: "Viewer SPA: dashboard flips between stats and 'new install' hero under live capture traffic (reload storm + failure!=empty conflation)" + severity: med + difficulty: med + kind: bug + note: "Viewer conflates fetch failure with empty, flipping between stats and the new-install hero." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1340 + fix_prs: [1342] + - number: 1341 + title: "mem::summarize: dead noop gate causes failure storm when no LLM key is set" + severity: med + difficulty: low + kind: bug + note: "Dead noop gate causes a summarize failure storm when no LLM key is set." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1341 + mentioned_by_prs: [1342] + - number: 1360 + title: "Oversized store fields and unpaginated full-list endpoints: sourceSessionIds/sourceMemoryIds bloat + 16 MiB WS frame limit" + severity: high + difficulty: high + kind: bug + note: "Unbounded sourceSessionIds/sourceMemoryIds plus unpaginated list endpoints hit the 16 MiB frame limit." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1360 + - number: 1361 + title: "Windows hardening: cross-platform build script, test normalization, and audit split" + severity: med + difficulty: med + kind: chore + note: "Windows build script portability, test normalization and audit split." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/issues/1361 + fix_prs: [1362] + - number: 1364 + title: "Bug: /insights 500 Invocation stopped — trailing garbage bytes appended to mem:insights.bin" + severity: high + difficulty: high + kind: bug + note: "Trailing garbage bytes appended to mem:insights.bin make GET /insights fail with Invocation stopped." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1364 + - number: 1365 + title: "session-end transcript back-fill extracts no prompts from a Claude Code transcript" + severity: med + difficulty: low + kind: bug + note: "extractTranscriptPrompts matches no Claude Code transcript shape, so prompt backfill posts nothing." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1365 + fix_prs: [1366] + - number: 1370 + title: "[viewer] WebUI randomly disconnects, backend remains healthy, only page refresh fixes it" + severity: med + difficulty: med + kind: bug + note: "Viewer randomly loses its connection while the backend stays healthy; only a refresh recovers." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1370 + - number: 1372 + title: "Semantic search silently stops saving new memories once the store passes a size set by the embedding width, and a restart does not rebuild them" + severity: high + difficulty: high + kind: bug + note: "Semantic search silently stops saving once the serialized index passes the string-length ceiling." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1372 + - number: 1373 + title: "Semantic search never stores anything with most non-OpenAI embedding models unless the operator sets the dimension variable by hand" + severity: high + difficulty: low + kind: bug + note: "Non-OpenAI embedding models never store vectors unless the operator sets the dimension variable by hand." + disposition: already-fixed + fixed_by: "Dimensions resolve through resolveDimensions() (src/providers/embedding/openai.ts:82)." + url: https://github.com/rohitg00/agentmemory/issues/1373 + - number: 1374 + title: "Unbounded sourceObservationIds inflate a 4062-node graph to 192MB; every traversal drops the worker and takes /search down with it" + severity: high + difficulty: med + kind: bug + note: "Unbounded provenance inflates a 4062-node graph to 192MB; every traversal drops the worker." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1374 + - number: 1377 + title: "[mcp] Some MCP host bridges receive empty inputSchema from tools/list — parameterized tool calls fail" + severity: high + difficulty: med + kind: bug + note: "Some MCP host bridges receive an empty inputSchema from tools/list, so parameterized calls fail." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1377 + - number: 1378 + title: "Worker heartbeat death loop on large legacy graph corpus: any vector result triggers blocking expandFromChunks (plus DROP_STALE_INDEX full-rebuild variant)" + severity: high + difficulty: high + kind: bug + note: "Any vector result triggers a blocking expandFromChunks, putting the worker in a heartbeat death loop." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1378 + - number: 1379 + title: "Bug: iii Console install prompt repeats after accepting installation in v0.9.29" + severity: low + difficulty: low + kind: bug + note: "iii Console install prompt repeats because the installer and detection logic disagree." + disposition: wont-fix + wont_fix_reason: "no iii engine — this fork runs one in-process SQLite Engine (ADR 0001)" + url: https://github.com/rohitg00/agentmemory/issues/1379 + - number: 1381 + title: "persistGraphDelta can zero the graph snapshot on a transient read failure (readSnapshot fail-open)" + severity: high + difficulty: low + kind: bug + note: "persistGraphDelta treats a failed snapshot read as empty and writes the zeroed snapshot back." + disposition: already-fixed + fixed_by: "persistGraphDelta wraps loadSnapshot in try/catch and abandons the batch rather than writing an empty view (src/functions/graph.ts:751-759)." + url: https://github.com/rohitg00/agentmemory/issues/1381 + fix_prs: [1384] + - number: 1382 + title: "paginateFromSnapshot can report a total below the nodes it returns; truncated banner never fires" + severity: low + difficulty: low + kind: bug + note: "paginateFromSnapshot can report a total below the node count it returns; the truncated banner never fires." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1382 + fix_prs: [1385] + - number: 1383 + title: "viewer: \"Rebuild Graph\" fires a full corpus replay with no confirmation, in-flight guard, or result handling" + severity: med + difficulty: low + kind: bug + note: "Viewer Rebuild Graph fires a full corpus replay with no confirmation, in-flight guard or result handling." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/issues/1383 + fix_prs: [1386] diff --git a/docs/upstream-pr-triage.yaml b/docs/upstream-pr-triage.yaml new file mode 100644 index 0000000000..392b58b199 --- /dev/null +++ b/docs/upstream-pr-triage.yaml @@ -0,0 +1,2388 @@ +# Upstream orphan-PR triage — rohitg00/agentmemory +# +# Scope: the 168 open upstream PRs that reference no open upstream issue, and are +# therefore absent from docs/upstream-issue-triage.yaml. The other 123 open PRs are +# cross-linked from the issues they claim to fix in that file. +# +# This fork does not merge, rebase or cherry-pick from upstream (see CLAUDE.md). These +# rows are read for the defect and the diagnosis, not for the patch. +# +# kind — fix | feature | provider | host | deps | docs | test | ci | deploy +# +# relevance — worth acting on in THIS fork: +# high a real defect or a recall-quality gain in code this fork carries and runs +# med applies here, but narrower: one host, one provider path, or a smaller win +# low another host/provider/deploy target, dependency bumps, or code absent here +# +# in_fork — /. A low ratio usually means +# the PR lands in files this fork deleted (deploy/, README translations, iii engine +# config, dropped hosts) — read it before assuming it does not apply. +# paths lists only the tracked ones, capped at 10. +# +# disposition — what this fork intends to do: +# already-fixed verified fixed in this tree; see status +# wont-fix out of scope here; wont_fix_reason says why +# candidate applies here and is unclaimed — the working set +# +# status — present only where the claim was checked against this tree by reading the +# code. "fixed-here" means this fork already carries the fix; "present-here" means the +# defect reproduces here as reported; "partly-present-here" means part of it was already +# fixed. Absent means unchecked: the note is still upstream's claim. +# +# Snapshot: 168 orphan PRs as of 2026-09-21 (gh pr list --repo rohitg00/agentmemory, +# changed files from repos/rohitg00/agentmemory/pulls/N/files). + +prs: + - number: 228 + title: "feat(vector): auto-index observations into vector store | 向量索引自动集成" + kind: feature + relevance: high + in_fork: 2/2 + churn: 19 + paths: + - src/functions/observe.ts + - src/index.ts + note: "Auto-index compressed observations into the vector store during observe; without it hybrid search needs a manual build." + disposition: already-fixed + status: "fixed-here — observe already indexes into the vector store via vectorIndexAddGuarded (src/functions/observe.ts:12,327)." + url: https://github.com/rohitg00/agentmemory/pull/228 + - number: 264 + title: "Implement surgical security and robustness improvements" + kind: fix + relevance: high + in_fork: 9/9 + churn: 110 + paths: + - src/auth.ts + - src/functions/claude-bridge.ts + - src/functions/snapshot.ts + - src/index.ts + - src/mcp/server.ts + - src/triggers/api.ts + - src/viewer/server.ts + - test/claude-bridge.test.ts + - test/snapshot.test.ts + note: "Viewer-proxy auth, snapshot write integrity and several hardening fixes across 9 files this fork carries." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/264 + - number: 300 + title: "feat: add sqlite-vec vector backend with Strategy pattern" + kind: feature + relevance: high + in_fork: 9/16 + churn: 805 + paths: + - benchmark/longmemeval-bench.ts + - benchmark/quality-eval.ts + - benchmark/real-embeddings-eval.ts + - benchmark/scale-eval.ts + - package.json + - src/index.ts + - src/state/hybrid-search.ts + - src/state/vector-index.ts + - test/vector-index.test.ts + note: "sqlite-vec vector backend behind a Strategy interface — the closest upstream work to this fork's own storage model." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/300 + - number: 470 + title: "Add consistency guards for MCP tools and REST endpoint drift" + kind: test + relevance: high + in_fork: 1/1 + churn: 20 + paths: + - test/consistency.test.ts + note: "Guards that every getAllTools() entry has a server switch case and that the endpoint count matches — directly automates this fork's Consistency Rules." + disposition: already-fixed + status: "fixed-here — test/consistency.test.ts and test/tool-count-consistency.test.ts already guard tool and endpoint drift." + url: https://github.com/rohitg00/agentmemory/pull/470 + - number: 532 + title: "fix(mcp): route memory_recall through full search" + kind: fix + relevance: high + in_fork: 3/3 + churn: 389 + paths: + - src/mcp/standalone.ts + - test/mcp-standalone-proxy.test.ts + - test/mcp-standalone.test.ts + note: "Standalone memory_recall proxies to smart-search, so format and token_budget are ignored and advertised recall modes do not work." + disposition: already-fixed + status: "fixed-here — Standalone memory_recall calls /agentmemory/search and forwards format and token_budget (src/mcp/standalone.ts:205-217); only memory_smart_search goes to /smart-search." + url: https://github.com/rohitg00/agentmemory/pull/532 + - number: 571 + title: "feat(smart-search): boost title/narrative matches on 'who/what is X' queries" + kind: feature + relevance: high + in_fork: 3/3 + churn: 262 + paths: + - src/functions/smart-search.ts + - src/types.ts + - test/smart-search.test.ts + note: "Boosts title/narrative matches on 'who/what is X' queries, where BM25 currently ranks busy observations over the record that names the concept." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/571 + - number: 572 + title: "perf(summarize): skip mem::summarize when a fresh summary exists" + kind: fix + relevance: high + in_fork: 1/1 + churn: 38 + paths: + - src/functions/summarize.ts + note: "Stop hooks fire mem::summarize every turn and stack full LLM passes; skips when a fresh summary exists." + disposition: candidate + status: "present-here — Same evidence as issue #1062: mem::summarize never checks for an existing fresh summary (src/functions/summarize.ts:235-360)." + url: https://github.com/rohitg00/agentmemory/pull/572 + - number: 615 + title: "feat(context,smart-search): inject insights into session context and smart search results (#457)" + kind: feature + relevance: high + in_fork: 2/2 + churn: 79 + paths: + - src/functions/context.ts + - src/functions/smart-search.ts + note: "mem::context and smart-search load lessons but skip insights, so mem::reflect output is reachable only by explicit tool call." + disposition: candidate + status: "present-here — KV.insights (src/state/schema.ts:88) is written by reflect.ts and read only by diagnostics.ts and export-import.ts. mem::context loads slots, profile and lessons and never insights (src/functions/context.ts:73-161); smart-search has no reference either." + url: https://github.com/rohitg00/agentmemory/pull/615 + - number: 667 + title: "fix(cli): status uses /memories?count=true instead of /export" + kind: fix + relevance: high + in_fork: 1/3 + churn: 83 + paths: + - src/cli.ts + note: "agentmemory status calls unbounded /export to get two counts and times out on any grown install." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/667 + - number: 757 + title: "fix(compression): fallback to synthetic compression when memory provider is noop" + kind: fix + relevance: high + in_fork: 2/2 + churn: 146 + paths: + - src/functions/compress.ts + - test/auto-compress.test.ts + note: "Compression against the noop provider produces nothing; falls back to local synthetic compression so keyless installs still store something." + disposition: already-fixed + status: "fixed-here — compress.ts already degrades to buildSyntheticCompression on provider failure (src/functions/compress.ts:113-130)." + url: https://github.com/rohitg00/agentmemory/pull/757 + - number: 829 + title: "Add project isolation for agentmemory" + kind: feature + relevance: high + in_fork: 50/56 + churn: 1605 + paths: + - .env.example + - plugin/.mcp.copilot.json + - plugin/.mcp.json + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + # +40 more + note: "AGENTMEMORY_PROJECT_ISOLATION enforced across MCP, REST and search, including a cross-project leak in hybrid smart-search." + disposition: candidate + status: "present-here — no AGENTMEMORY_PROJECT_ISOLATION." + url: https://github.com/rohitg00/agentmemory/pull/829 + - number: 878 + title: "Security: escape stored memory content in injected context (H1)" + kind: fix + relevance: high + in_fork: 5/6 + churn: 316 + paths: + - src/functions/context.ts + - src/functions/enrich.ts + - src/functions/slots.ts + - src/prompts/xml.ts + - test/context-lessons.test.ts + note: "Stored memory content is interpolated raw into , so a memory can close the wrapper and inject instructions." + disposition: candidate + status: "present-here — src/functions/context.ts escapes only the project attribute (line 26); stored fields are still interpolated raw." + url: https://github.com/rohitg00/agentmemory/pull/878 + - number: 898 + title: "fix(viewer): prevent path-traversal SSRF in proxy (CWE-918)" + kind: fix + relevance: high + in_fork: 1/2 + churn: 221 + paths: + - src/viewer/server.ts + note: "Path-traversal SSRF (CWE-918) in the viewer proxy: local callers escape the /agentmemory/ prefix with the auth header attached." + disposition: candidate + status: "present-here — pathname is a raw slice of req.url with no normalization (src/viewer/server.ts:253) and the proxy concatenates it straight into the upstream URL (src/viewer/server.ts:399-403) while attaching Authorization: Bearer . A ../ segment resolves during fetch, so the secret rides a request outside the /agentmemory/ prefix. Local-only reach." + url: https://github.com/rohitg00/agentmemory/pull/898 + - number: 922 + title: "fix(security): close secret-scrubbing bypass paths in all MCP write routes" + kind: fix + relevance: high + in_fork: 8/11 + churn: 1599 + paths: + - src/functions/compress.ts + - src/functions/export-import.ts + - src/functions/lessons.ts + - src/functions/privacy.ts + - src/functions/remember.ts + - src/functions/sketches.ts + - src/functions/slots.ts + - src/functions/team.ts + note: "stripPrivateData runs only in the observe pipeline; seven MCP write tools store secrets in plaintext." + disposition: candidate + status: "present-here — stripPrivateData is called only from src/functions/observe.ts." + url: https://github.com/rohitg00/agentmemory/pull/922 + - number: 928 + title: "feat(search): add temporal decay to recall ranking" + kind: feature + relevance: high + in_fork: 4/8 + churn: 1676 + paths: + - .env.example + - src/config.ts + - src/index.ts + - src/state/hybrid-search.ts + note: "Opt-in exponential temporal decay in RRF fusion; today a note from this morning ranks identically to one from a year ago." + disposition: candidate + status: "present-here — no temporal decay in src/state/hybrid-search.ts." + url: https://github.com/rohitg00/agentmemory/pull/928 + - number: 975 + title: "Fix: unified environment behavior" + kind: fix + relevance: high + in_fork: 18/19 + churn: 210 + paths: + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/session-start.mjs + - plugin/scripts/stop.mjs + # +8 more + note: "Every hook script reads AGENTMEMORY_* at module level, so config living only in ~/.agentmemory/.env is silently ignored; touches 18 hook scripts this fork carries." + disposition: candidate + status: "present-here — No hook imports ../config.js; pre-tool-use.ts has no imports at all and reads process.env at module load (src/hooks/pre-tool-use.ts:23-26) while its own comment tells you to set the value in ~/.agentmemory/.env (line 20). Only the daemon hydrates that file (src/config.ts:75-83), so hook-only config set there is silently ignored." + url: https://github.com/rohitg00/agentmemory/pull/975 + - number: 1005 + title: "fix: evicted observations leave ghost entries in search indexes" + kind: fix + relevance: high + in_fork: 1/1 + churn: 13 + paths: + - src/functions/evict.ts + note: "Evicted observations are never removed from the BM25 or vector index, so search returns ids whose KV lookup is null and silently under-fills pages." + disposition: candidate + status: "present-here — src/functions/evict.ts deletes from KV without touching the BM25 or vector index." + url: https://github.com/rohitg00/agentmemory/pull/1005 + - number: 1039 + title: "fix: schedule periodic snapshot timer (#1006)" + kind: fix + relevance: high + in_fork: 1/2 + churn: 93 + paths: + - src/index.ts + note: "SNAPSHOT_ENABLED is loaded and logged but no timer ever calls mem::snapshot-create — the feature never runs." + disposition: already-fixed + status: "fixed-here — the periodic snapshot timer exists (src/index.ts:357)." + url: https://github.com/rohitg00/agentmemory/pull/1039 + - number: 1174 + title: "fix: evict lowest-value observations at the session cap instead of refusing writes" + kind: fix + relevance: high + in_fork: 1/2 + churn: 211 + paths: + - src/functions/observe.ts + note: "At MAX_OBS_PER_SESSION every further observation is discarded; evicts the lowest-value one instead of dropping the newest." + disposition: candidate + status: "present-here — At MAX_OBS_PER_SESSION mem::observe refuses every further observation outright (src/functions/observe.ts:147-154). The newest work is dropped rather than the least valuable one evicted." + url: https://github.com/rohitg00/agentmemory/pull/1174 + - number: 1175 + title: "feat(slots): reject stale slot overwrites, keep an undo copy, report headroom" + kind: fix + relevance: high + in_fork: 4/5 + churn: 389 + paths: + - src/functions/slots.ts + - src/mcp/server.ts + - src/mcp/tools-registry.ts + - src/types.ts + note: "mem::slot-replace is an unguarded whole-object write, so two sessions silently clobber each other." + disposition: candidate + status: "partly-present-here — mem::slot-replace is serialized by withKeyedLock (src/functions/slots.ts:319), so it is not the unguarded write described. It still has no version precondition and no undo copy, so a read-modify-write across two sessions silently loses the first write." + url: https://github.com/rohitg00/agentmemory/pull/1175 + - number: 1176 + title: "fix: bound the audit log with AGENTMEMORY_AUDIT_MAX" + kind: fix + relevance: high + in_fork: 1/2 + churn: 143 + paths: + - src/functions/audit.ts + note: "recordAudit has no retention and the KV adapter rewrites the whole scope per write, so audit cost compounds on long-lived installs." + disposition: candidate + status: "present-here — recordAudit has no retention bound." + url: https://github.com/rohitg00/agentmemory/pull/1176 + - number: 1210 + title: "fix(providers): replace retired Anthropic default model identifiers" + kind: fix + relevance: high + in_fork: 3/3 + churn: 16 + paths: + - .env.example + - src/config.ts + - src/providers/index.ts + note: "claude-sonnet-4-20250514 was retired on 15 June 2026 and is still the fallback default in six places." + disposition: already-fixed + status: "fixed-here — claude-sonnet-4-20250514 appears nowhere in src/." + url: https://github.com/rohitg00/agentmemory/pull/1210 + - number: 1220 + title: "fix(hooks): scope pre-tool-use enrichment to the current project" + kind: fix + relevance: high + in_fork: 2/3 + churn: 199 + paths: + - plugin/scripts/pre-tool-use.mjs + - src/hooks/pre-tool-use.ts + note: "PreToolUse carries no project field, so hook enrichment is always unscoped and pulls context across projects." + disposition: already-fixed + status: "fixed-here — pre-tool-use resolves project and forwards it on the enrich call (src/hooks/pre-tool-use.ts:97-111)." + url: https://github.com/rohitg00/agentmemory/pull/1220 + - number: 1247 + title: "feat(api): add /agentmemory/memories/search" + kind: feature + relevance: high + in_fork: 1/1 + churn: 179 + paths: + - src/triggers/api.ts + note: "No search endpoint covers memories written by remember — /memories only lists." + disposition: candidate + status: "present-here — no /agentmemory/memories/search route in src/triggers/api.ts." + url: https://github.com/rohitg00/agentmemory/pull/1247 + - number: 1259 + title: "fix(providers): bound LLM provider concurrency and stop counting 429s as breaker failures" + kind: fix + relevance: high + in_fork: 1/3 + churn: 341 + paths: + - src/providers/resilient.ts + note: "No concurrency bound on the provider path and 429s count as circuit-breaker failures, so load trips the breaker." + disposition: candidate + status: "partly-present-here — 429 and 503 are already retried with Retry-After before reaching the breaker (src/providers/_fetch.ts:13,76-82). A concurrency cap exists, but only on the agent-sdk provider (src/providers/agent-sdk.ts:99) — the HTTP providers have none." + url: https://github.com/rohitg00/agentmemory/pull/1259 + - number: 1260 + title: "fix(mcp): read graph stats from the snapshot instead of enumerating the graph" + kind: fix + relevance: high + in_fork: 2/2 + churn: 56 + paths: + - src/mcp/server.ts + - test/mcp-resources.test.ts + note: "agentmemory://graph/stats enumerates every node and edge over the transport to compute four aggregates." + disposition: already-fixed + status: "fixed-here — mem::graph-stats reads the snapshot and never enumerates (src/functions/graph.ts:1105)." + url: https://github.com/rohitg00/agentmemory/pull/1260 + - number: 1287 + title: "fix(evict): schedule the eviction sweep at boot" + kind: fix + relevance: high + in_fork: 5/6 + churn: 204 + paths: + - .env.example + - src/config.ts + - src/functions/evict.ts + - src/index.ts + - test/evict.test.ts + note: "mem::evict is fully implemented but nothing schedules it — Eviction never runs." + disposition: candidate + status: "present-here — mem::evict is registered (src/functions/evict.ts:109) but src/index.ts schedules only auto-forget and consolidate-pipeline." + url: https://github.com/rohitg00/agentmemory/pull/1287 + - number: 1307 + title: "fix(sessions): close idle sessions no client ever ended" + kind: fix + relevance: high + in_fork: 5/11 + churn: 862 + paths: + - src/config.ts + - src/index.ts + - src/triggers/api.ts + - src/triggers/events.ts + - src/types.ts + note: "Sessions only reach completed when a client POSTs /session/end; clients that never do leave sessions active forever." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1307 + - number: 1309 + title: "feat(graph): delta graph extraction" + kind: fix + relevance: high + in_fork: 5/5 + churn: 250 + paths: + - src/functions/graph.ts + - src/state/schema.ts + - test/graph-heuristic-extract.test.ts + - test/graph.test.ts + - test/schema.test.ts + note: "Graph extraction re-runs over all observations every trigger, duplicating nodes and burning LLM calls." + disposition: candidate + status: "present-here — no graphExtracted scope in src/state/schema.ts." + url: https://github.com/rohitg00/agentmemory/pull/1309 + - number: 1310 + title: "fix(compression): guard empty compression prompts and watermark summarization" + kind: fix + relevance: high + in_fork: 10/11 + churn: 575 + paths: + - src/functions/compress-synthetic.ts + - src/functions/compress.ts + - src/functions/summarize.ts + - src/prompts/compression.ts + - src/state/schema.ts + - src/triggers/api.ts + - src/triggers/events.ts + - src/types.ts + - test/schema.test.ts + - test/summarize.test.ts + note: "Empty observations produce empty compression prompts, and summarization re-runs on an unchanged observation count." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1310 + - number: 1317 + title: "fix(health): measure heap saturation against the heap ceiling, not heapTotal" + kind: fix + relevance: high + in_fork: 4/4 + churn: 62 + paths: + - src/health/monitor.ts + - src/health/thresholds.ts + - src/types.ts + - test/health-thresholds.test.ts + note: "Health measured heapUsed/heapTotal, which reads 80-95% in any healthy process; measures against the real ceiling instead." + disposition: already-fixed + status: "fixed-here — saturation is measured against the heap ceiling (src/health/thresholds.ts:88)." + url: https://github.com/rohitg00/agentmemory/pull/1317 + - number: 1319 + title: "fix(consolidation): dedupe double-fired consolidation LLM runs + crash-safe audit" + kind: fix + relevance: high + in_fork: 3/3 + churn: 383 + paths: + - src/functions/audit.ts + - src/functions/consolidation-pipeline.ts + - test/consolidation-pipeline.test.ts + note: "The consolidation pipeline fires the same full-corpus LLM run twice for one corpus state." + disposition: already-fixed + status: "fixed-here — The consolidation pipeline already holds a lock that joins a second run over the same corpus state (src/functions/consolidation-pipeline.ts:52-56)." + url: https://github.com/rohitg00/agentmemory/pull/1319 + - number: 1324 + title: "fix(viewer): dashboard shows 0 sessions when unbounded KV scopes overflow iii invocation" + kind: fix + relevance: high + in_fork: 2/2 + churn: 58 + paths: + - src/triggers/api.ts + - src/viewer/index.html + note: "api::semantic-list and siblings run unbounded kv.list over a ~17MB scope, overflowing the invocation and blanking the dashboard." + disposition: candidate + status: "present-here — api::semantic-list and api::procedural-list run unbounded kv.list with no limit and return the whole scope (src/triggers/api.ts:2235, 2249)." + url: https://github.com/rohitg00/agentmemory/pull/1324 + - number: 1344 + title: "feat(reflect): project-level namespace isolation for reflection and consolidation" + kind: fix + relevance: high + in_fork: 6/6 + churn: 191 + paths: + - src/functions/consolidation-pipeline.ts + - src/functions/reflect.ts + - src/mcp/server.ts + - src/mcp/tools-registry.ts + - test/consolidation-pipeline.test.ts + - test/reflect.test.ts + note: "Reflection and consolidation are not project-scoped, so concept clusters and insights cross-contaminate between repos." + disposition: candidate + status: "partly-present-here — mem::reflect accepts and filters on project (src/functions/reflect.ts:187,354,385), so the capability is present; it is opt-in, and the scheduled run passes no project." + url: https://github.com/rohitg00/agentmemory/pull/1344 + - number: 1350 + title: "fix(providers): clear the circuit breaker failure count on success" + kind: fix + relevance: high + in_fork: 4/4 + churn: 138 + paths: + - src/config.ts + - src/providers/circuit-breaker.ts + - src/providers/resilient.ts + - test/circuit-breaker.test.ts + note: "recordSuccess() does not reset the failure count in the closed state, so unrelated failures accumulate until the breaker trips." + disposition: candidate + status: "partly-present-here — recordSuccess only resets in half-open (src/providers/circuit-breaker.ts:47-54), but recordFailure already zeroes the count when the previous failure is older than the 60s window (src/providers/circuit-breaker.ts:63-65). Failures do not accumulate forever; the real gap is that successes never reduce the count inside a window." + url: https://github.com/rohitg00/agentmemory/pull/1350 + - number: 1351 + title: "perf(search): reuse one graph view per store and bound traversal" + kind: fix + relevance: high + in_fork: 3/5 + churn: 703 + paths: + - src/functions/graph-retrieval.ts + - src/state/hybrid-search.ts + - src/state/kv.ts + note: "Rebuilds the graph view per search and traverses unbounded; reuses one view per store and caps the traversal." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1351 + - number: 1367 + title: "fix(summarize): detect wrapped noop provider so keyless installs skip the pipeline" + kind: fix + relevance: high + in_fork: 2/2 + churn: 27 + paths: + - src/functions/summarize.ts + - test/summarize.test.ts + note: "The noop provider is wrapped, so the no_provider early return never fires and keyless installs run the whole summarize pipeline on every Stop hook." + disposition: candidate + status: "present-here — Every provider is wrapped: createProvider returns ResilientProvider (src/providers/index.ts:55) whose name is `resilient()` (src/providers/resilient.ts:9). summarize.ts:263 compares name === \"noop\", which can never match, so keyless installs run the whole summarize pipeline. graph.ts:926 already uses .includes(\"noop\") — the fix has a precedent in this tree." + url: https://github.com/rohitg00/agentmemory/pull/1367 + - number: 1368 + title: "fix(reflect): read bounded graph snapshot and cap cluster prompt" + kind: fix + relevance: high + in_fork: 6/6 + churn: 326 + paths: + - .env.example + - plugin/skills/agentmemory-config/REFERENCE.md + - src/functions/graph.ts + - src/functions/reflect.ts + - src/prompts/reflect.ts + - test/reflect.test.ts + note: "mem::reflect enumerates the full graph and blows the invocation limit on a large store; bounds the snapshot and caps the cluster prompt." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1368 + - number: 218 + title: "fix(viewer): prevent tab bar crush and fix graph canvas overflow" + kind: fix + relevance: med + in_fork: 1/1 + churn: 12 + paths: + - src/viewer/index.html + note: "Two viewer CSS bugs: tab bar collapses to ~8px and the graph canvas overflows." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/218 + - number: 223 + title: "feat(smart-search): add format parameter (full/narrative/compact) | 添加 format 参数" + kind: feature + relevance: med + in_fork: 1/1 + churn: 48 + paths: + - src/functions/smart-search.ts + note: "format=full/narrative/compact on mem::smart-search, matching mem::search." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/223 + - number: 224 + title: "feat(search): add CJK character bigram tokenization | 添加 CJK Bigram 分词" + kind: feature + relevance: med + in_fork: 1/1 + churn: 17 + paths: + - src/state/search-index.ts + note: "CJK bigram tokenization in the BM25 index; without it CJK text is effectively unsearchable." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/224 + - number: 225 + title: "feat(embedding): configurable local model, cache dir, and HF mirror | 嵌入模型可配置" + kind: feature + relevance: med + in_fork: 1/1 + churn: 38 + paths: + - src/providers/embedding/local.ts + note: "Env-configurable local embedding model, cache dir and HF mirror." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/225 + - number: 266 + title: "fix: suppress floating promise in keyed-mutex to prevent unhandled-rejection warnings" + kind: fix + relevance: med + in_fork: 1/1 + churn: 6 + paths: + - src/state/keyed-mutex.ts + note: "Floating promise in withKeyedLock can raise unhandled-rejection warnings." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/266 + - number: 273 + title: "feat: opt-in fail-loud when AGENTMEMORY_URL is unreachable" + kind: fix + relevance: med + in_fork: 2/4 + churn: 131 + paths: + - src/mcp/rest-proxy.ts + - test/mcp-standalone-proxy.test.ts + note: "The MCP shim silently falls back to local in-memory mode when AGENTMEMORY_URL is unreachable; this makes it opt-in loud." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/273 + - number: 319 + title: "fix(fs-watcher): load agentmemory env file" + kind: fix + relevance: med + in_fork: 3/3 + churn: 194 + paths: + - integrations/filesystem-watcher/README.md + - integrations/filesystem-watcher/watcher.mjs + - test/fs-watcher.test.ts + note: "fs-watcher never loads ~/.agentmemory/.env, so its config is silently ignored." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/319 + - number: 365 + title: "fix: tolerate partial dashboard payloads" + kind: fix + relevance: med + in_fork: 1/2 + churn: 278 + paths: + - src/viewer/index.html + note: "Dashboard crashes on partial payloads instead of rendering empty lists." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/365 + - number: 465 + title: "feat(connect): automate Hermes config merge" + kind: host + relevance: med + in_fork: 3/3 + churn: 299 + paths: + - src/cli/connect/hermes.ts + - src/cli/connect/util.ts + - test/cli-connect.test.ts + note: "Turns the Hermes connect stub into a real ~/.hermes/config.yaml merge." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/465 + - number: 467 + title: "fix(embeddings): allow custom OpenAI embedding path" + kind: fix + relevance: med + in_fork: 4/5 + churn: 57 + paths: + - .env.example + - src/providers/_openai-shared.ts + - src/providers/embedding/openai.ts + - test/embedding-provider.test.ts + note: "OPENAI_EMBEDDING_PATH for OpenAI-compatible endpoints that do not use /v1/embeddings." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/467 + - number: 468 + title: "fix(memory): audit remember attribution" + kind: fix + relevance: med + in_fork: 8/9 + churn: 279 + paths: + - src/functions/remember.ts + - src/mcp/server.ts + - src/mcp/standalone.ts + - src/mcp/tools-registry.ts + - src/types.ts + - test/mcp-standalone-proxy.test.ts + - test/mcp-standalone.test.ts + - test/remember-forget-audit.test.ts + note: "mem::remember records no audit row and drops sessionId/tags from attribution." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/468 + - number: 502 + title: "fix: run graph extraction on manual session end" + kind: fix + relevance: med + in_fork: 2/3 + churn: 180 + paths: + - src/triggers/api.ts + - src/triggers/events.ts + note: "POST /session/end skips the stopped-session pipeline, so graph extraction never runs on a manual end." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/502 + - number: 529 + title: "fix(viewer): show explicit error banner when dashboard endpoints fail" + kind: fix + relevance: med + in_fork: 8/9 + churn: 227 + paths: + - plugin/hooks/hooks.codex.json + - plugin/hooks/hooks.json + - src/config.ts + - src/index.ts + - src/mcp/standalone.ts + - src/viewer/index.html + - test/env-loader.test.ts + - test/mcp-standalone-proxy.test.ts + note: "Dashboard shows 0 memories rather than an error when the backend is down." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/529 + - number: 568 + title: "feat(graph): auto-trigger mem::graph-extract on mem::remember" + kind: feature + relevance: med + in_fork: 2/2 + churn: 34 + paths: + - src/functions/remember.ts + - test/cross-project-isolation.test.ts + note: "mem::remember never triggers graph-extract, so explicit saves never enter the graph." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/568 + - number: 569 + title: "feat(provider): AGENTMEMORY_DISABLE_THINKING to recover structured output on Qwen3/GLM/Kimi" + kind: feature + relevance: med + in_fork: 3/3 + churn: 32 + paths: + - .env.example + - plugin/skills/agentmemory-config/REFERENCE.md + - src/providers/openai.ts + note: "AGENTMEMORY_DISABLE_THINKING to recover structured output from hybrid-reasoning models (Qwen3/GLM/Kimi)." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/569 + - number: 610 + title: "fix: iii reconnect route loss + Voyage env config + heap calc against actual ceiling" + kind: fix + relevance: med + in_fork: 3/3 + churn: 413 + paths: + - src/health/thresholds.ts + - src/index.ts + - src/providers/embedding/voyage.ts + note: "Three unrelated fixes; the heap-ceiling one duplicates #1317 and the iii reconnect one does not apply here." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/610 + - number: 665 + title: "fix: pass agentId per-call for integration-scoped memory" + kind: fix + relevance: med + in_fork: 6/8 + churn: 214 + paths: + - integrations/hermes/__init__.py + - integrations/openclaw/plugin.mjs + - src/config.ts + - src/mcp/server.ts + - test/agent-id-scope.test.ts + - test/openclaw-plugin.test.ts + note: "Integrations send no agentId, so every memory lands unscoped even with multi-agent isolation on." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/665 + - number: 689 + title: "feat(team): per-request userId + TEAM_MODE=private enforcement" + kind: fix + relevance: med + in_fork: 5/7 + churn: 393 + paths: + - src/config.ts + - src/functions/team.ts + - src/mcp/server.ts + - src/mcp/tools-registry.ts + - test/team.test.ts + note: "TEAM_MODE=private was read but never enforced, and team-share accepted any sharedBy value." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/689 + - number: 742 + title: "fix: LLM provider works with subscription proxies + honor SUMMARIZE_CHUNK_SIZE from .env" + kind: fix + relevance: med + in_fork: 2/2 + churn: 19 + paths: + - src/functions/summarize.ts + - src/providers/anthropic.ts + note: "Folds the Anthropic system prompt into messages for subscription proxies, and honors SUMMARIZE_CHUNK_SIZE." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/742 + - number: 802 + title: "fix: prefer embedding API key for OpenAI embeddings" + kind: fix + relevance: med + in_fork: 2/2 + churn: 23 + paths: + - src/providers/embedding/index.ts + - test/embedding-provider.test.ts + note: "createEmbeddingProvider passes OPENAI_API_KEY over a set OPENAI_EMBEDDING_API_KEY; same defect as #832." + disposition: already-fixed + status: "fixed-here — OPENAI_EMBEDDING_API_KEY takes precedence (src/providers/embedding/openai.ts:61)." + url: https://github.com/rohitg00/agentmemory/pull/802 + - number: 832 + title: "Fix OpenAI embedding provider key selection" + kind: fix + relevance: med + in_fork: 2/2 + churn: 16 + paths: + - src/providers/embedding/index.ts + - test/embedding-provider.test.ts + note: "Prefer OPENAI_EMBEDDING_API_KEY in the embedding factory; cleaner duplicate of #802." + disposition: already-fixed + status: "fixed-here — same as #802." + url: https://github.com/rohitg00/agentmemory/pull/832 + - number: 846 + title: "Add stdio MCP resources and prompts" + kind: feature + relevance: med + in_fork: 2/2 + churn: 556 + paths: + - src/mcp/standalone.ts + - test/mcp-standalone.test.ts + note: "Resources and prompts capabilities on the stdio MCP shim, proxied to the running server." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/846 + - number: 862 + title: "Allow XENOVA_CACHE_HOME to redirect local embedding model path" + kind: fix + relevance: med + in_fork: 1/1 + churn: 18 + paths: + - src/providers/embedding/local.ts + note: "XENOVA_CACHE_HOME to redirect the local embedding model path out of node_modules." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/862 + - number: 927 + title: "fix(cli): add clear warning when import-jsonl brings zero files" + kind: fix + relevance: med + in_fork: 1/1 + churn: 6 + paths: + - src/cli.ts + note: "import-jsonl reports success when it imported zero files." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/927 + - number: 953 + title: "fix: preserve Zed settings during connect" + kind: fix + relevance: med + in_fork: 5/5 + churn: 146 + paths: + - package.json + - src/cli/connect/json-mcp-adapter.ts + - src/cli/connect/util.ts + - src/cli/connect/zed.ts + - test/connect-new-agents.test.ts + note: "connect zed overwrites existing settings.json instead of merging JSONC." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/953 + - number: 970 + title: "fix: make consolidation timeouts configurable" + kind: fix + relevance: med + in_fork: 4/13 + churn: 270 + paths: + - .env.example + - src/config.ts + - src/functions/consolidate.ts + - src/index.ts + note: "Consolidate per-concept and worker invocation timeouts hardcoded; adjacent to this fork's own LLM-timeout work." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/970 + - number: 974 + title: "fix(context,compress-synthetic): conversation observations render with full content" + kind: fix + relevance: med + in_fork: 5/6 + churn: 102 + paths: + - .github/workflows/ci.yml + - package.json + - src/functions/compress-synthetic.ts + - src/functions/context.ts + - website/package.json + note: "tool_name 'conversation' observations render as type 'other' with no content in injected context." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/974 + - number: 1011 + title: "feat(replay): support OpenAI chat-format JSONL in parseJsonlText" + kind: feature + relevance: med + in_fork: 2/3 + churn: 129 + paths: + - src/replay/jsonl-parser.ts + - test/replay.test.ts + note: "parseJsonlText only knows the Claude Code transcript shape; OpenAI chat-format JSONL imports produce zero observations." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1011 + - number: 1023 + title: "fix(replay): surface compressed observation content in replay detail" + kind: fix + relevance: med + in_fork: 2/2 + churn: 153 + paths: + - src/functions/replay.ts + - test/replay.test.ts + note: "Replay detail panel is empty because rawFromCompressed() drops the compressed content." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1023 + - number: 1037 + title: "fix: CJK-aware remember dedup with negation guard (#1021)" + kind: fix + relevance: med + in_fork: 2/5 + churn: 395 + paths: + - src/functions/remember.ts + - src/state/schema.ts + note: "remember dedup tokenizes on whitespace, so repeated CJK preferences never dedupe." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1037 + - number: 1089 + title: "feat: forward agentId on remember + Hermes multi-profile project tagging" + kind: fix + relevance: med + in_fork: 12/14 + churn: 1064 + paths: + - integrations/hermes/README.md + - integrations/hermes/__init__.py + - src/config.ts + - src/functions/remember.ts + - src/mcp/server.ts + - src/mcp/standalone.ts + - src/mcp/tools-registry.ts + - src/triggers/api.ts + - src/triggers/events.ts + - test/agent-id-scope.test.ts + # +2 more + note: "remember and memory_save drop agentId, and standalone MCP drops project despite advertising it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1089 + - number: 1177 + title: "feat(health): surface search-index freshness" + kind: feature + relevance: med + in_fork: 3/5 + churn: 189 + paths: + - src/health/monitor.ts + - src/health/thresholds.ts + - src/types.ts + note: "Health surfaces search-index freshness; today a broken persister is visible only as a throttled warn line." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1177 + - number: 1215 + title: "fix(embedding): route transformers cacheDir to user home cache" + kind: fix + relevance: med + in_fork: 5/7 + churn: 210 + paths: + - __mocks__/@huggingface/transformers.ts + - src/providers/embedding/clip.ts + - src/providers/embedding/local.ts + - src/state/reranker.ts + - test/local-embedding-provider.test.ts + note: "transformers writes its model cache next to its own package, which fails on global or containerized installs." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1215 + - number: 1290 + title: "fix(metrics): window failure rate to recent activity + track lastFailureAt" + kind: fix + relevance: med + in_fork: 3/4 + churn: 125 + paths: + - src/eval/metrics-store.ts + - src/health/monitor.ts + - src/types.ts + note: "Health reports a cumulative failure rate, so a fixed bug leaves a stale high number forever." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1290 + - number: 1291 + title: "fix(metrics): serialize MetricsStore.record per function" + kind: fix + relevance: med + in_fork: 1/2 + churn: 130 + paths: + - src/eval/metrics-store.ts + note: "MetricsStore.record() loses updates: concurrent callers meet in the kv.get() gap." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1291 + - number: 1293 + title: "fix(hooks): retry a failed observe POST instead of dropping it" + kind: fix + relevance: med + in_fork: 16/18 + churn: 654 + paths: + - plugin/scripts/notification.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/subagent-start.mjs + - plugin/scripts/subagent-stop.mjs + - plugin/scripts/task-completed.mjs + - src/hooks/notification.ts + - src/hooks/post-tool-failure.ts + # +6 more + note: "Adds retry/queue to hook observe POSTs; conflicts with this fork's fire-and-forget hook contract, so take the diagnosis, not the design." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1293 + - number: 1308 + title: "fix(pipeline): classify telemetry observations and extract semantic fields for non-tool events" + kind: fix + relevance: med + in_fork: 8/13 + churn: 1416 + paths: + - plugin/opencode/agentmemory-capture.ts + - src/functions/compress-synthetic.ts + - src/functions/consolidate.ts + - src/functions/context.ts + - src/functions/observe.ts + - src/functions/summarize.ts + - src/prompts/summary.ts + - src/types.ts + note: "Non-tool hook events carry payload data that mem::observe never extracts, so telemetry observations are contentless." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1308 + - number: 1311 + title: "fix(opencode): skip replayed history observation on session fork" + kind: host + relevance: med + in_fork: 1/3 + churn: 348 + paths: + - plugin/opencode/agentmemory-capture.ts + note: "Forking an OpenCode session re-observes the whole conversation history as new observations." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1311 + - number: 1318 + title: "fix(replay): reinforce lesson confidence on JSONL re-import" + kind: fix + relevance: med + in_fork: 2/3 + churn: 99 + paths: + - src/functions/lessons.ts + - src/functions/replay.ts + note: "JSONL re-import bumps reinforcements but never confidence, unlike reinforceLesson()." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1318 + - number: 1322 + title: "fix(memory): reject leaked tool-call args in remember" + kind: fix + relevance: med + in_fork: 1/2 + churn: 185 + paths: + - src/functions/remember.ts + note: "A mis-encoded tool call writes its trailing arguments into memory content as literal text." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1322 + - number: 1323 + title: "fix(mcp): stop masking tool errors as unknown-tool fallback" + kind: fix + relevance: med + in_fork: 4/4 + churn: 57 + paths: + - src/mcp/rest-proxy.ts + - src/mcp/server.ts + - src/mcp/standalone.ts + - test/mcp-standalone-proxy.test.ts + note: "Any non-ok proxy response is treated as unreachable, hiding real tool errors behind an unknown-tool fallback." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1323 + - number: 1327 + title: "fix(clip): encode text with CLIP text tower so text→image vision-search works" + kind: fix + relevance: med + in_fork: 2/2 + churn: 96 + paths: + - src/providers/embedding/clip.ts + - test/clip-embedding-provider.test.ts + note: "ClipEmbeddingProvider encoded text with the image tower, so text-to-image vision-search always failed." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1327 + - number: 1336 + title: "fix(slots): line-boundary truncation in slot reflection to prevent mid-line slicing" + kind: fix + relevance: med + in_fork: 34/51 + churn: 7795 + paths: + - .gitignore + - plugin/opencode/agentmemory-capture.ts + - src/functions/audit.ts + - src/functions/compress-synthetic.ts + - src/functions/compress.ts + - src/functions/consolidation-pipeline.ts + - src/functions/diagnostics.ts + - src/functions/graph.ts + - src/functions/observe.ts + - src/functions/reflect.ts + # +24 more + note: "Line-boundary truncation in slot reflection instead of a raw mid-line slice; the fix is small but the PR carries 7.8k churn across 51 files." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1336 + - number: 1363 + title: "feat(scope): dynamic project identity and tiered observation lifecycle" + kind: feature + relevance: med + in_fork: 57/100 + churn: 13022 + paths: + - .gitignore + - CONTEXT.md + - docs/agents/domain.md + - docs/agents/issue-tracker.md + - plugin/opencode/agentmemory-capture.ts + - plugin/scripts/notification.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/prompt-submit.mjs + # +47 more + note: "Git-remote-derived canonical project slugs and a tiered observation lifecycle; the right-scope idea is good, the 13k-churn PR is not." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1363 + - number: 1369 + title: "fix(embeddings): send OPENAI_EMBEDDING_DIMENSIONS as the dimensions field" + kind: fix + relevance: med + in_fork: 3/4 + churn: 218 + paths: + - .env.example + - src/providers/embedding/openai.ts + - test/embedding-provider.test.ts + note: "OPENAI_EMBEDDING_DIMENSIONS is read into the provider but never sent in the request body, breaking matryoshka models." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1369 + - number: 194 + title: "feat: add --host flag and AGENTMEMORY_HOST for configurable bind address" + kind: feature + relevance: low + in_fork: 6/6 + churn: 114 + paths: + - src/auth.ts + - src/cli.ts + - src/config.ts + - src/index.ts + - src/types.ts + - src/viewer/server.ts + note: "--host/AGENTMEMORY_HOST to bind off localhost; the fork already runs one local daemon." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/194 + - number: 219 + title: "feat(providers): add Azure OpenAI provider" + kind: provider + relevance: low + in_fork: 4/5 + churn: 109 + paths: + - src/config.ts + - src/providers/index.ts + - src/types.ts + - src/viewer/index.html + note: "Azure OpenAI LLM provider." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/219 + - number: 220 + title: "fix(providers): support Azure AI Foundry (Anthropic) endpoint and corporate proxy" + kind: provider + relevance: low + in_fork: 21/25 + churn: 579 + paths: + - package.json + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/session-start.mjs + # +11 more + note: "Azure OpenAI / AI Foundry plus HTTP CONNECT corporate-proxy support; overlaps #594." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/220 + - number: 226 + title: "feat(health): configurable health check thresholds via env vars | 健康阈值可配置" + kind: feature + relevance: low + in_fork: 1/1 + churn: 14 + paths: + - src/health/thresholds.ts + note: "Health thresholds via env; the defaults false-alarm on small hosts." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/226 + - number: 227 + title: "feat(shutdown): graceful shutdown timeout with force exit | 优雅关闭超时" + kind: fix + relevance: low + in_fork: 1/1 + churn: 7 + paths: + - src/index.ts + note: "Force-exit if graceful shutdown hangs on the iii WebSocket — a failure mode this fork does not have." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/227 + - number: 241 + title: "feat: hierarchical outline index for long markdown artifacts" + kind: feature + relevance: low + in_fork: 7/16 + churn: 1328 + paths: + - plugin/.claude-plugin/plugin.json + - src/index.ts + - src/mcp/server.ts + - src/mcp/tools-registry.ts + - src/state/schema.ts + - src/triggers/api.ts + - src/types.ts + note: "PageIndex-style outline index, 3 MCP tools + 3 endpoints; a new subsystem." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/241 + - number: 245 + title: "feat: add Codex Desktop hooks integration" + kind: host + relevance: low + in_fork: 0/4 + churn: 635 + note: "Codex Desktop hook bridge; Codex is not a supported host here." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/245 + - number: 267 + title: "Enhance auto-forget with redundancy tracking" + kind: feature + relevance: low + in_fork: 1/1 + churn: 76 + paths: + - src/functions/auto-forget.ts + note: "Redundancy tracking in auto-forget; CodeRabbit-authored, thin rationale." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/267 + - number: 302 + title: "fix(openclaw): register memory capability for memory slot" + kind: host + relevance: low + in_fork: 2/2 + churn: 19 + paths: + - integrations/openclaw/README.md + - integrations/openclaw/plugin.mjs + note: "OpenClaw memory-slot capability registration." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/302 + - number: 333 + title: "Refactor health monitoring logic and snapshot collection" + kind: chore + relevance: low + in_fork: 1/1 + churn: 61 + paths: + - src/health/monitor.ts + note: "Health monitor refactor, 5s cadence; CodeRabbit-authored." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/333 + - number: 376 + title: "feat: configurable bind hosts for iii-engine and viewer" + kind: feature + relevance: low + in_fork: 2/3 + churn: 47 + paths: + - src/cli.ts + - src/viewer/server.ts + note: "Configurable bind host for engine and viewer; overlaps #194." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/376 + - number: 378 + title: "build(docker): unify per-platform deploys behind single GHCR image" + kind: deploy + relevance: low + in_fork: 0/16 + churn: 793 + note: "Single GHCR image for all platform deploys; no file here." + disposition: wont-fix + wont_fix_reason: "deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/pull/378 + - number: 464 + title: "docs: refresh integration capability counts" + kind: docs + relevance: low + in_fork: 6/8 + churn: 62 + paths: + - .env.example + - assets/tags/light/stat-tests.svg + - assets/tags/stat-tests.svg + - integrations/hermes/README.md + - integrations/openclaw/README.md + - test/consistency.test.ts + note: "Refreshes stale tool/test counts in integration docs." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/464 + - number: 488 + title: "fix(hermes): keep plugin hook manifest in sync" + kind: fix + relevance: low + in_fork: 2/2 + churn: 57 + paths: + - integrations/hermes/plugin.yaml + - test/consistency.test.ts + note: "Hermes plugin.yaml missing hook names, plus a manifest consistency guard." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/488 + - number: 528 + title: "fix(cli): tie iii engine lifecycle and skip prompts in non-TTY context" + kind: fix + relevance: low + in_fork: 8/9 + churn: 195 + paths: + - plugin/hooks/hooks.codex.json + - plugin/hooks/hooks.json + - src/cli.ts + - src/config.ts + - src/index.ts + - src/mcp/standalone.ts + - test/env-loader.test.ts + - test/mcp-standalone-proxy.test.ts + note: "TTY heuristic orphans iii.exe; no iii engine in this fork." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/528 + - number: 543 + title: "Improve Hermes manual install status messaging" + kind: chore + relevance: low + in_fork: 2/2 + churn: 25 + paths: + - src/cli/connect/hermes.ts + - src/viewer/index.html + note: "Replaces the Hermes 'stub' install status with clearer wording." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/543 + - number: 558 + title: "Add Azure OpenAI provider configuration" + kind: provider + relevance: low + in_fork: 9/11 + churn: 285 + paths: + - .env.example + - src/cli.ts + - src/cli/doctor-diagnostics.ts + - src/cli/onboarding.ts + - src/config.ts + - src/functions/summarize.ts + - src/providers/index.ts + - src/providers/openai.ts + - test/openai-shared.test.ts + note: "Azure OpenAI config path; third competing Azure PR with #219/#594." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/558 + - number: 566 + title: "docs: sync MCP tool counts across metadata" + kind: docs + relevance: low + in_fork: 4/5 + churn: 37 + paths: + - .env.example + - plugin/.claude-plugin/plugin.json + - src/mcp/standalone.ts + - website/components/Compare.tsx + note: "Tool-count sync across metadata." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/566 + - number: 583 + title: "Add optional external_id metadata passthrough for memory retrieval results" + kind: feature + relevance: low + in_fork: 8/11 + churn: 219 + paths: + - src/functions/remember.ts + - src/functions/search.ts + - src/mcp/server.ts + - src/mcp/standalone.ts + - src/mcp/tools-registry.ts + - src/state/memory-utils.ts + - src/triggers/api.ts + - src/types.ts + note: "external_id + metadata passthrough for imported memories." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/583 + - number: 594 + title: "feat(providers): add Azure OpenAI / AI Foundry provider with HTTP CONNECT proxy support" + kind: provider + relevance: low + in_fork: 20/21 + churn: 354 + paths: + - package.json + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/session-start.mjs + # +10 more + note: "Azure OpenAI / AI Foundry with proxy support; overlaps #220/#558." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/594 + - number: 604 + title: "feat(pi): add full lifecycle hooks to pi integration" + kind: host + relevance: low + in_fork: 1/1 + churn: 246 + paths: + - integrations/pi/index.ts + note: "pi integration from 3 to 6 lifecycle hooks." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/604 + - number: 606 + title: "feat(providers): add DeepSeek provider" + kind: provider + relevance: low + in_fork: 2/4 + churn: 106 + paths: + - src/providers/index.ts + - src/types.ts + note: "DeepSeek embedding provider." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/606 + - number: 607 + title: "feat(cli): allow AGENTMEMORY_IMPORT_TIMEOUT_MS override for import-jsonl" + kind: fix + relevance: low + in_fork: 1/1 + churn: 20 + paths: + - src/cli.ts + note: "AGENTMEMORY_IMPORT_TIMEOUT_MS instead of a hardcoded 120s import ceiling." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/607 + - number: 617 + title: "feat: add privacy-safe audit receipts" + kind: feature + relevance: low + in_fork: 6/7 + churn: 196 + paths: + - src/functions/audit.ts + - src/mcp/server.ts + - src/mcp/standalone.ts + - src/mcp/tools-registry.ts + - src/triggers/api.ts + - test/audit.test.ts + note: "Hashed privacy-safe audit receipt format." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/617 + - number: 624 + title: "fix(cli): honor --force on unresponsive engine; detect listening PIDs on Windows" + kind: fix + relevance: low + in_fork: 1/2 + churn: 183 + paths: + - src/cli.ts + note: "Windows: --force ignored on a wedged engine socket." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/624 + - number: 630 + title: "feat(connect): add Kilo CLI adapter" + kind: host + relevance: low + in_fork: 1/2 + churn: 114 + paths: + - src/cli/connect/index.ts + note: "Kilo CLI connect adapter." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/630 + - number: 634 + title: "fix(windows): make agentmemory actually start on Windows (watch path + IPv4 engine WS + configurable timeout)" + kind: fix + relevance: low + in_fork: 2/4 + churn: 44 + paths: + - src/cli.ts + - src/config.ts + note: "Windows startup failure rooted in the iii engine watch path and WS binding." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/634 + - number: 672 + title: "feat(viewer): add timeline sort toggle" + kind: feature + relevance: low + in_fork: 2/2 + churn: 127 + paths: + - src/viewer/index.html + - test/viewer-session-id.test.ts + note: "Viewer timeline newest/oldest sort toggle." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/672 + - number: 679 + title: "feat: add memory metacognition foundation" + kind: feature + relevance: low + in_fork: 4/15 + churn: 2864 + paths: + - src/index.ts + - src/state/schema.ts + - src/triggers/api.ts + - src/types.ts + note: "Metacognition phase 1: policy, write candidates, readback, 8 endpoints." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/679 + - number: 693 + title: "fix(cli): prioritize user iii config lookup" + kind: fix + relevance: low + in_fork: 3/7 + churn: 190 + paths: + - .gitignore + - package.json + - src/cli.ts + note: "iii-config lookup precedence; no iii engine here." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/693 + - number: 694 + title: "fix(viewer): preserve dashboard scroll and show API errors" + kind: feature + relevance: low + in_fork: 2/2 + churn: 190 + paths: + - src/viewer/index.html + - test/viewer-session-id.test.ts + note: "Preserve dashboard scroll across refreshes plus API error toasts." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/694 + - number: 699 + title: "feat(cli): materialize ~/.agentmemory/config/iii-config.yaml with absolute paths" + kind: fix + relevance: low + in_fork: 1/4 + churn: 378 + paths: + - src/cli.ts + note: "Materializes ~/.agentmemory/config/iii-config.yaml with absolute paths." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/699 + - number: 711 + title: "feat: add AGENTMEMORY_OUTPUT_LANG to control generated-text language" + kind: feature + relevance: low + in_fork: 2/3 + churn: 74 + paths: + - .env.example + - src/providers/resilient.ts + note: "AGENTMEMORY_OUTPUT_LANG for generated memory text; overlaps #914." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/711 + - number: 721 + title: "fix(viewer): correct Token Savings calculation to match CLI" + kind: fix + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - src/viewer/index.html + note: "Viewer token-savings formula disagrees with the CLI and reads 0%." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/721 + - number: 741 + title: "feat: AWS Bedrock provider (LLM + embeddings) with SSO + auth-refresh" + kind: provider + relevance: low + in_fork: 9/16 + churn: 1338 + paths: + - .env.example + - package.json + - src/config.ts + - src/functions/compress-file.ts + - src/providers/embedding/index.ts + - src/providers/index.ts + - src/providers/resilient.ts + - src/types.ts + - test/compress-file.test.ts + note: "AWS Bedrock LLM + embeddings with SigV4 and SSO auth refresh." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/741 + - number: 756 + title: "feat(embedding): support configurable vector dimensions for OpenRouter provider" + kind: feature + relevance: low + in_fork: 2/2 + churn: 65 + paths: + - src/providers/embedding/openrouter.ts + - test/embedding-provider.test.ts + note: "OPENROUTER_EMBEDDING_DIMENSIONS." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/756 + - number: 799 + title: "fix: handle undefined manifest value in loadShardedData" + kind: fix + relevance: low + in_fork: 0/1 + churn: 2 + note: "Undefined-vs-null manifest check crashes the sharded loader on upgrade; the file does not exist in this fork." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/799 + - number: 818 + title: "feat: upgrade MiniMax default model to M3" + kind: chore + relevance: low + in_fork: 8/8 + churn: 20 + paths: + - .env.example + - src/cli/onboarding.ts + - src/config.ts + - src/providers/index.ts + - src/providers/minimax.ts + - test/fallback-model-resolution.test.ts + - test/fetch-timeout.test.ts + - test/minimax-provider.test.ts + note: "MiniMax default model to M3; duplicates #1056." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/818 + - number: 822 + title: "fix(codex): declare agentmemory MCP server as stdio" + kind: host + relevance: low + in_fork: 0/2 + churn: 79 + note: "type = stdio in the Codex MCP block." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/822 + - number: 841 + title: "feat(deploy): add systemd service template" + kind: deploy + relevance: low + in_fork: 0/3 + churn: 324 + note: "systemd unit template; no deploy/ tree in this fork." + disposition: wont-fix + wont_fix_reason: "deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/pull/841 + - number: 860 + title: "[codex] Fix OpenClaw terminal capture" + kind: host + relevance: low + draft: true + in_fork: 2/2 + churn: 614 + paths: + - integrations/openclaw/plugin.mjs + - test/openclaw-plugin.test.ts + note: "DRAFT. OpenClaw terminal capture through llm_output/agent_end." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/860 + - number: 891 + title: "Hermes plugin: stable project scoping + close sessions on process exit" + kind: fix + relevance: low + in_fork: 1/1 + churn: 114 + paths: + - integrations/hermes/__init__.py + note: "Hermes project scoping under ephemeral worktrees, plus session close on exit." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/891 + - number: 914 + title: "Add configurable prompt output language" + kind: feature + relevance: low + in_fork: 6/8 + churn: 136 + paths: + - .env.example + - plugin/skills/agentmemory-config/REFERENCE.md + - src/prompts/compression.ts + - src/prompts/graph-extraction.ts + - src/prompts/reflect.ts + - src/prompts/summary.ts + note: "AGENTMEMORY_OUTPUT_LANGUAGE across compression/summary/graph/reflect prompts; overlaps #711." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/914 + - number: 920 + title: "feat(connect): add Qoder CLI CN adapter" + kind: host + relevance: low + in_fork: 1/5 + churn: 451 + paths: + - src/cli/connect/index.ts + note: "Qoder CLI CN connect adapter." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/920 + - number: 923 + title: "fly: Enhance Dockerfile with local embeddings installation" + kind: deploy + relevance: low + in_fork: 0/1 + churn: 7 + note: "Fly Dockerfile preloads local embedding weights." + disposition: wont-fix + wont_fix_reason: "deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/pull/923 + - number: 963 + title: "feat(providers): add configurable OpenRouter and Gemini base URLs" + kind: feature + relevance: low + in_fork: 8/12 + churn: 311 + paths: + - .env.example + - src/config.ts + - src/providers/embedding/gemini.ts + - src/providers/embedding/openrouter.ts + - src/providers/index.ts + - src/providers/openrouter.ts + - test/fallback-model-resolution.test.ts + - test/fetch-timeout.test.ts + note: "OPENROUTER_BASE_URL and GEMINI_BASE_URL, and splits Gemini into its own provider." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/963 + - number: 967 + title: "ci(deps)(deps): bump actions/checkout from 6 to 7" + kind: deps + relevance: low + in_fork: 1/2 + churn: 4 + paths: + - .github/workflows/ci.yml + note: "dependabot: actions/checkout 6 to 7." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/967 + - number: 1000 + title: "Add wire support for IBM Bob" + kind: host + relevance: low + in_fork: 1/3 + churn: 17 + paths: + - src/cli/connect/index.ts + note: "IBM Bob connect adapter." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1000 + - number: 1012 + title: "feat(hermes): capture tool calls individually and remove content truncation in sync_turn" + kind: fix + relevance: low + in_fork: 2/2 + churn: 123 + paths: + - integrations/hermes/__init__.py + - test/hermes-plugin.test.ts + note: "Hermes sync_turn truncates tool_input/tool_output and collapses tool calls." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1012 + - number: 1016 + title: "Support Codex pre-tool-use targets" + kind: host + relevance: low + draft: true + in_fork: 5/6 + churn: 433 + paths: + - plugin/hooks/hooks.codex.json + - plugin/scripts/pre-tool-use.mjs + - src/hooks/pre-tool-use.ts + - test/codex-connect-hooks.test.ts + - test/context-injection.test.ts + note: "DRAFT. Codex-native PreToolUse payload adaptation." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1016 + - number: 1049 + title: "feat: add Amp (Sourcegraph) integration with auto-capture plugin" + kind: host + relevance: low + in_fork: 3/8 + churn: 782 + paths: + - package.json + - src/cli/connect/index.ts + - test/cli-connect.test.ts + note: "Amp (Sourcegraph) integration with auto-capture plugin." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1049 + - number: 1053 + title: "deps(website)(deps-dev): bump typescript from 6.0.3 to 7.0.2 in /website" + kind: deps + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - website/package.json + note: "dependabot: website typescript 6 to 7." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1053 + - number: 1055 + title: "feat: durable candidates MVP + scoped traceable recall + Codex archive watcher" + kind: feature + relevance: low + in_fork: 51/92 + churn: 7774 + paths: + - .env.example + - .gitignore + - package.json + - plugin/plugin.json + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + # +41 more + note: "Durable candidates MVP, traceable recall and a Codex archive watcher; 7.7k churn across 92 files." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1055 + - number: 1056 + title: "Update MiniMax provider defaults for MiniMax-M3" + kind: chore + relevance: low + in_fork: 7/8 + churn: 112 + paths: + - .env.example + - src/cli/onboarding.ts + - src/config.ts + - src/providers/index.ts + - src/providers/minimax.ts + - test/fallback-model-resolution.test.ts + - test/minimax-provider.test.ts + note: "MiniMax M3 defaults plus base-URL normalization; supersedes #1032, duplicates #818." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1056 + - number: 1076 + title: "docs(hermes): enable copied provider plugin" + kind: docs + relevance: low + draft: true + in_fork: 1/2 + churn: 10 + paths: + - integrations/hermes/README.md + note: "DRAFT. Documents the hermes plugins enable step." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1076 + - number: 1082 + title: "ci(deps)(deps): bump actions/setup-node from 6 to 7" + kind: deps + relevance: low + in_fork: 1/2 + churn: 4 + paths: + - .github/workflows/ci.yml + note: "dependabot: actions/setup-node 6 to 7." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1082 + - number: 1084 + title: "deps(deps-dev): bump typescript from 6.0.3 to 7.0.2" + kind: deps + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - package.json + note: "dependabot: typescript 6 to 7." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1084 + - number: 1085 + title: "deps(deps-dev): bump @types/node from 25.9.5 to 26.1.1" + kind: deps + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - package.json + note: "dependabot: @types/node 25 to 26." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1085 + - number: 1111 + title: "feat(providers): add Cloudflare Workers AI as an LLM and embedding provider" + kind: provider + relevance: low + in_fork: 12/17 + churn: 676 + paths: + - .env.example + - src/cli.ts + - src/cli/doctor-diagnostics.ts + - src/cli/onboarding.ts + - src/config.ts + - src/functions/consolidation-pipeline.ts + - src/functions/summarize.ts + - src/providers/embedding/index.ts + - src/providers/index.ts + - src/types.ts + # +2 more + note: "Cloudflare Workers AI as LLM and embedding provider." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1111 + - number: 1112 + title: "feat(cursor): workspace adapter on top of the marketplace plugin" + kind: host + relevance: low + in_fork: 2/20 + churn: 3972 + paths: + - .gitignore + - tsdown.config.ts + note: "Cursor workspace adapter; Cursor is not supported here." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1112 + - number: 1139 + title: "fix: make build and test suite pass on Windows" + kind: fix + relevance: low + in_fork: 6/7 + churn: 136 + paths: + - package.json + - test/cli-remove.test.ts + - test/compress-file.test.ts + - test/copilot-plugin.test.ts + - test/obsidian-export.test.ts + - test/slots-flag-gate.test.ts + note: "Cross-platform build script and Windows test fixes." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/1139 + - number: 1147 + title: "fix(docs): sync REST endpoint counts + skills reference (unblocks CI on main)" + kind: docs + relevance: low + in_fork: 2/4 + churn: 8 + paths: + - plugin/skills/agentmemory-rest-api/REFERENCE.md + - src/index.ts + note: "Endpoint-count sync to unblock upstream CI." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1147 + - number: 1150 + title: "fix(cli): refresh Codex hooks when MCP is already wired" + kind: host + relevance: low + in_fork: 3/4 + churn: 169 + paths: + - plugin/skills/agentmemory-agents/REFERENCE.md + - plugin/skills/agentmemory-rest-api/REFERENCE.md + - test/codex-connect-hooks.test.ts + note: "connect codex --with-hooks skips hook install when MCP is already wired." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1150 + - number: 1164 + title: "feat: add docker-compose deployment and config for local embeddings" + kind: deploy + relevance: low + in_fork: 2/8 + churn: 229 + paths: + - src/index.ts + - src/providers/embedding/local.ts + note: "docker-compose deployment plus configurable local embedding model/dimensions." + disposition: wont-fix + wont_fix_reason: "deploy target this fork does not carry" + url: https://github.com/rohitg00/agentmemory/pull/1164 + - number: 1172 + title: "docs: document Novita AI as an OpenAI-compatible LLM provider" + kind: docs + relevance: low + in_fork: 7/9 + churn: 241 + paths: + - .env.example + - src/config.ts + - src/providers/_openai-shared.ts + - src/providers/index.ts + - src/providers/openai.ts + - test/fallback-model-resolution.test.ts + - test/openai-shared.test.ts + note: "Documents Novita AI as an OpenAI-compatible endpoint." + disposition: wont-fix + wont_fix_reason: "upstream doc/count sync or dependency bump" + url: https://github.com/rohitg00/agentmemory/pull/1172 + - number: 1185 + title: "fix(pi): scope memory_save memories to the current project" + kind: host + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - integrations/pi/index.ts + note: "pi memory_save had no project, so its memories landed global." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1185 + - number: 1186 + title: "feat(pi): summarize and end the session on quit" + kind: host + relevance: low + in_fork: 1/1 + churn: 30 + paths: + - integrations/pi/index.ts + note: "pi never ended sessions, leaving zombies with no summary." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1186 + - number: 1206 + title: "feat(pi): observe per-tool results (PostToolUse parity)" + kind: host + relevance: low + in_fork: 1/1 + churn: 41 + paths: + - integrations/pi/index.ts + note: "pi per-tool observations for PostToolUse parity." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1206 + - number: 1207 + title: "feat(pi): robustness — client dedup, session/start, prompt_submit, larger turn capture" + kind: host + relevance: low + in_fork: 1/1 + churn: 52 + paths: + - integrations/pi/index.ts + note: "pi client-side dedup, session/start, prompt_submit, larger turn capture." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1207 + - number: 1222 + title: "feat: klaatcode support (cli adapter, native hooks)" + kind: host + relevance: low + in_fork: 19/25 + churn: 559 + paths: + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/session-start.mjs + - plugin/scripts/subagent-start.mjs + - plugin/scripts/subagent-stop.mjs + # +9 more + note: "Klaat Code connect adapter and native hooks." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1222 + - number: 1227 + title: "fix(pi): attribute sessions to Pi cwd" + kind: host + relevance: low + in_fork: 2/2 + churn: 15 + paths: + - integrations/pi/index.ts + - test/connect-pi.test.ts + note: "pi attributes sessions to ctx.cwd." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1227 + - number: 1234 + title: "fix: reject state reads without keys" + kind: fix + relevance: low + in_fork: 1/2 + churn: 23 + paths: + - src/state/kv.ts + note: "StateKV rejects missing/empty keys before hitting the engine." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1234 + - number: 1248 + title: "feat: expose hook REST endpoints as MCP tools" + kind: feature + relevance: low + in_fork: 2/2 + churn: 407 + paths: + - src/mcp/server.ts + - src/mcp/tools-registry.ts + note: "Exposes 6 hook REST endpoints as MCP tools for hostless MCP clients." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1248 + - number: 1257 + title: "Project: Windows USB portable kit (in-tree + runtime pack)" + kind: feature + relevance: low + in_fork: 14/36 + churn: 1506 + paths: + - .gitignore + - plugin/scripts/notification.mjs + - plugin/scripts/post-commit.mjs + - plugin/scripts/post-tool-failure.mjs + - plugin/scripts/post-tool-use.mjs + - plugin/scripts/pre-compact.mjs + - plugin/scripts/pre-tool-use.mjs + - plugin/scripts/prompt-submit.mjs + - plugin/scripts/session-end.mjs + - plugin/scripts/session-start.mjs + # +4 more + note: "Windows USB portable kit under agentmemory-portable/." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/1257 + - number: 1261 + title: "fix(test): point USERPROFILE at the temp home on Windows" + kind: test + relevance: low + in_fork: 1/1 + churn: 12 + paths: + - test/slots-flag-gate.test.ts + note: "Windows: point USERPROFILE at the temp home." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/1261 + - number: 1262 + title: "fix(test): normalise fs mock paths so they match on Windows" + kind: test + relevance: low + in_fork: 2/2 + churn: 35 + paths: + - test/compress-file.test.ts + - test/obsidian-export.test.ts + note: "Windows: normalize fs mock paths." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/1262 + - number: 1263 + title: "fix(test): drop POSIX-only assumptions from six suites" + kind: test + relevance: low + in_fork: 2/6 + churn: 30 + paths: + - test/cli-remove.test.ts + - test/copilot-plugin.test.ts + note: "Windows: drop POSIX assumptions from six suites." + disposition: wont-fix + wont_fix_reason: "Windows — CI runs ubuntu and macos only" + url: https://github.com/rohitg00/agentmemory/pull/1263 + - number: 1301 + title: "fix(opencode): dynamic project repository detection from tool execution and workspace paths" + kind: host + relevance: low + in_fork: 1/2 + churn: 511 + paths: + - plugin/opencode/agentmemory-capture.ts + note: "OpenCode resolves the project directory once at startup and stays locked to it." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1301 + - number: 1320 + title: "feat: Kimi Code CLI plugin support" + kind: host + relevance: low + in_fork: 0/3 + churn: 103 + note: "Kimi Code CLI plugin manifest." + disposition: wont-fix + wont_fix_reason: "no connect adapter for this host in src/cli/connect/" + url: https://github.com/rohitg00/agentmemory/pull/1320 + - number: 1355 + title: "feat: add Requesty provider (chat + embeddings)" + kind: provider + relevance: low + in_fork: 17/32 + churn: 469 + paths: + - .env.example + - src/cli.ts + - src/cli/doctor-diagnostics.ts + - src/cli/onboarding.ts + - src/config.ts + - src/functions/consolidation-pipeline.ts + - src/functions/summarize.ts + - src/providers/embedding/index.ts + - src/providers/index.ts + - src/types.ts + # +7 more + note: "Requesty router as LLM and embedding provider." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1355 + - number: 1356 + title: "mcp-http: per-request agent attribution for the shared bridge" + kind: feature + relevance: low + in_fork: 57/100 + churn: 18717 + paths: + - .env.example + - .github/workflows/ci.yml + - assets/tags/light/stat-tools.svg + - assets/tags/stat-tools.svg + - benchmark/README.md + - integrations/hermes/README.md + - integrations/hermes/__init__.py + - integrations/openclaw/README.md + - integrations/openclaw/plugin.mjs + - integrations/pi/README.md + # +47 more + note: "Per-request agent attribution on the shared Streamable HTTP MCP bridge; 18.7k churn across 100 files." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1356 + - number: 1357 + title: "fix(cli): stop interpolating $HOME-derived paths into the iii installer shell command" + kind: fix + relevance: low + in_fork: 1/2 + churn: 125 + paths: + - src/cli.ts + note: "Shell injection through $HOME-derived paths in the iii installer command; this fork has no iii installer." + disposition: wont-fix + wont_fix_reason: "no iii engine here" + url: https://github.com/rohitg00/agentmemory/pull/1357 + - number: 1375 + title: "feat: add OrcaRouter as a first-class provider with API-key and OAuth 2.0 + PKCE login" + kind: provider + relevance: low + in_fork: 16/33 + churn: 6469 + paths: + - .env.example + - .gitignore + - package.json + - plugin/skills/agentmemory-rest-api/REFERENCE.md + - src/cli.ts + - src/cli/onboarding.ts + - src/config.ts + - src/index.ts + - src/providers/embedding/index.ts + - src/providers/index.ts + # +6 more + note: "OrcaRouter as a first-class provider with OAuth 2.0 + PKCE login." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1375 + - number: 1376 + title: "feat(providers): dedicated opencode provider for OpenCode Go relay" + kind: provider + relevance: low + in_fork: 4/7 + churn: 246 + paths: + - src/config.ts + - src/providers/index.ts + - src/providers/openai.ts + - src/types.ts + note: "Dedicated provider for the OpenCode Go relay, which requires an x-opencode-session header." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1376 + - number: 1380 + title: "deps(website)(deps-dev): bump @types/node from 25.9.8 to 26.6.1 in /website" + kind: deps + relevance: low + in_fork: 1/1 + churn: 2 + paths: + - website/package.json + note: "dependabot: website @types/node 25 to 26." + disposition: candidate + url: https://github.com/rohitg00/agentmemory/pull/1380 diff --git a/src/cli.ts b/src/cli.ts index 772694b4e6..7cbe79d26d 100644 --- a/src/cli.ts +++ b/src/cli.ts @@ -511,7 +511,7 @@ async function runStatus() { try { const [healthRes, sessionsRes, graphRes, memoriesRes, flagsRes, followupRes] = await Promise.all([ apiFetch(base, "health"), - apiFetch(base, "sessions"), + apiFetch(base, "sessions?limit=all"), apiFetch(base, "graph/stats"), apiFetch(base, "memories?count=true"), apiFetch(base, "config/flags"), diff --git a/src/engine/inproc/sdk.ts b/src/engine/inproc/sdk.ts index 70d6b434fc..057817838d 100644 --- a/src/engine/inproc/sdk.ts +++ b/src/engine/inproc/sdk.ts @@ -251,8 +251,10 @@ export function createInprocSdk(opts: InprocSdkOptions): InprocSdk { let streamBytes = 0; const subscriptions = new Set(); + // Separator is U+001F, not NUL: a literal NUL in the source makes git and + // grep classify this whole file as binary and silently skip it in searches. const groupKey = (streamName: string, groupId: string) => - `${streamName}${groupId}`; + `${streamName}${groupId}`; function broadcast(message: Record): void { const streamName = message.streamName as string; diff --git a/src/functions/search.ts b/src/functions/search.ts index 8c28884f21..2d7b9bf43a 100644 --- a/src/functions/search.ts +++ b/src/functions/search.ts @@ -761,15 +761,17 @@ export function registerSearchFunction(sdk: ISdk, kv: StateKV): void { if (!tokenBudget) return { items, used: items.reduce((sum, item) => sum + estimateTokens(item), 0), truncated: false } const selected: T[] = [] let used = 0 + let truncated = false for (const item of items) { const itemTokens = estimateTokens(item) if (used + itemTokens > tokenBudget) { - return { items: selected, used, truncated: selected.length < items.length } + truncated = true + continue } selected.push(item) used += itemTokens } - return { items: selected, used, truncated: false } + return { items: selected, used, truncated } } if (format === 'compact') { diff --git a/src/providers/_fetch.ts b/src/providers/_fetch.ts index 3b02d9352d..b90f55c747 100644 --- a/src/providers/_fetch.ts +++ b/src/providers/_fetch.ts @@ -3,13 +3,10 @@ import { getEnvVar } from "../config.js"; // Bounded retry for transient rate-limit / unavailable responses. Attempts is // total tries (initial + retries). Retries are bounded by a TOTAL elapsed // deadline — not per-attempt — so the worst case never blows past the caller's -// timeout budget or the iii invocation timeout. A single retry delay is capped -// low so a hostile Retry-After header can't dominate the budget. +// timeout budget. A single retry delay is capped low so a hostile Retry-After +// header can't dominate the budget. const MAX_ATTEMPTS = 3; const MAX_RETRY_DELAY_MS = 5000; -// Absolute ceiling on the total budget, kept well under the iii 180s invocation -// timeout so retries + sleeps + per-attempt timeouts can never overrun it. -const HARD_BUDGET_CAP_MS = 170000; // A retry only makes sense if there's room for at least a token attempt after // the sleep; without this floor we'd sleep, fire, and get instantly cut off. const MIN_ATTEMPT_FLOOR_MS = 100; @@ -65,15 +62,14 @@ export async function fetchWithTimeout( Number.parseInt(getEnvVar("AGENTMEMORY_LLM_TIMEOUT_MS") ?? "60000", 10); const ms = Number.isFinite(parsed) && parsed > 0 ? parsed : 60000; - // The caller's timeout is the TOTAL budget for all attempts + sleeps, hard - // capped so we never approach the iii invocation timeout. - const budgetMs = Math.min(ms, HARD_BUDGET_CAP_MS); + // The caller's timeout is the TOTAL budget for all attempts + sleeps, and it + // is honored exactly. This fork runs only the in-process Engine (ADR 0001), + // which imposes no invocation timeout to stay under, so a configured bound + // must not be silently shortened — a slow local model legitimately needs the + // minutes it was given. const start = Date.now(); - // The first attempt must honor the capped budget too — passing raw `ms` - // here would let a large caller timeout hang past HARD_BUDGET_CAP_MS (and - // the iii 180s invocation timeout) before any retry logic runs. - let response: Response = await fetchOnce(url, init, budgetMs); + let response: Response = await fetchOnce(url, init, ms); for (let attempt = 1; attempt < MAX_ATTEMPTS; attempt++) { if (!RETRY_STATUS.has(response.status)) return response; @@ -86,7 +82,7 @@ export async function fetchWithTimeout( // budget — a hostile Retry-After that alone exceeds the remaining budget // returns the last response instead of stalling the caller. const elapsed = Date.now() - start; - const remaining = budgetMs - elapsed; + const remaining = ms - elapsed; if (delay + MIN_ATTEMPT_FLOOR_MS > remaining) return response; // This response is being discarded for a retry; release its body so the @@ -98,7 +94,7 @@ export async function fetchWithTimeout( // can't push total elapsed past the deadline. const attemptMs = Math.max( MIN_ATTEMPT_FLOOR_MS, - Math.min(ms, budgetMs - (Date.now() - start)), + ms - (Date.now() - start), ); response = await fetchOnce(url, init, attemptMs); } diff --git a/src/triggers/api.ts b/src/triggers/api.ts index 08ff40089b..081bcac2aa 100644 --- a/src/triggers/api.ts +++ b/src/triggers/api.ts @@ -138,6 +138,32 @@ function parseOptionalPositiveInt(value: unknown): number | undefined | null { return parsed; } +const DEFAULT_PAGE_LIMIT = 100; + +type Page = { limit: number | "all"; offset: number }; + +function parsePage(query: Record | undefined): Page { + const rawLimit = query?.["limit"]; + const rawOffset = query?.["offset"]; + const parsedLimit = typeof rawLimit === "string" ? Number(rawLimit) : Number.NaN; + const parsedOffset = typeof rawOffset === "string" ? Number(rawOffset) : Number.NaN; + const limit = + rawLimit === "all" + ? "all" + : Number.isInteger(parsedLimit) && parsedLimit > 0 + ? parsedLimit + : DEFAULT_PAGE_LIMIT; + const offset = + Number.isInteger(parsedOffset) && parsedOffset >= 0 ? parsedOffset : 0; + return { limit, offset }; +} + +function takePage(rows: T[], page: Page): T[] { + return page.limit === "all" + ? rows.slice(page.offset) + : rows.slice(page.offset, page.offset + page.limit); +} + export function registerApiTriggers( sdk: ISdk, kv: StateKV, @@ -665,6 +691,11 @@ export function registerApiTriggers( body: { error: "sessionId is required and must be a non-empty string" }, }; } + // kv.update creates a missing key, so a stop for a Session that never + // started would invent a row; refuse it instead. + if (!(await kv.get(KV.sessions, sessionId))) { + return { status_code: 404, body: { error: "session_not_found" } }; + } await kv.update(KV.sessions, sessionId, [ { type: "set", path: "endedAt", value: new Date().toISOString() }, { type: "set", path: "status", value: "completed" }, @@ -862,17 +893,21 @@ export function registerApiTriggers( ? undefined : explicitAgentId ?? (isAgentScopeIsolated() ? getAgentId() : undefined); - const filtered = filterAgentId - ? sessions.filter((s) => s.agentId === filterAgentId) - : sessions; + const filtered = ( + filterAgentId + ? sessions.filter((s) => s.agentId === filterAgentId) + : sessions + ).sort((a, b) => (b.startedAt ?? "").localeCompare(a.startedAt ?? "")); + const page = parsePage(req.query_params); + const paged = takePage(filtered, page); // Bounded fan-out: each kv.get is a full engine invocation, so // Promise.all over hundreds of sessions saturates the invocation // pool. Batch in chunks of 10 (parallel within a chunk, sequential // across chunks); the summaries array stays index-aligned with - // `filtered`. + // `paged`. const summaries: Array = []; - for (let batch = 0; batch < filtered.length; batch += 10) { - const chunk = filtered.slice(batch, batch + 10); + for (let batch = 0; batch < paged.length; batch += 10) { + const chunk = paged.slice(batch, batch + 10); const results = await Promise.all( chunk.map((s) => kv.get(KV.summaries, s.id).catch(() => null), @@ -880,10 +915,13 @@ export function registerApiTriggers( ); summaries.push(...results); } - const withSummary = filtered.map((s, i) => + const withSummary = paged.map((s, i) => summaries[i] ? { ...s, summary: summaries[i] } : s, ); - return { status_code: 200, body: { sessions: withSummary } }; + return { + status_code: 200, + body: { sessions: withSummary, total: filtered.length, ...page }, + }; }, ); sdk.registerTrigger({ @@ -2130,13 +2168,8 @@ export function registerApiTriggers( ); } - // viewer + `agentmemory status` were hitting this endpoint to - // count memories. On a real corpus (8K+ memories) the unbounded - // response either timed out at the iii engine boundary ("Invocation - // stopped") or arrived too large for the viewer to render — so the - // UI showed 0 memories despite a healthy store. Two opt-in modes: - // ?count=true — totals only, no payload - // ?limit=N&offset=M — page slice (default unlimited for back-compat) + // ?count=true answers totals only, so the viewer and + // `agentmemory status` never pull the rows just to count them. if (req.query_params?.["count"] === "true") { // Match the SAME scope that the list path applies — returning // unfiltered totals here would leak cross-agent counts to a @@ -2150,28 +2183,13 @@ export function registerApiTriggers( }; } - const rawLimit = req.query_params?.["limit"]; - const rawOffset = req.query_params?.["offset"]; - const parsedLimit = - typeof rawLimit === "string" ? Number(rawLimit) : Number.NaN; - const parsedOffset = - typeof rawOffset === "string" ? Number(rawOffset) : Number.NaN; - const limit = - Number.isInteger(parsedLimit) && parsedLimit > 0 - ? Math.min(parsedLimit, 5000) - : undefined; - const offset = - Number.isInteger(parsedOffset) && parsedOffset >= 0 ? parsedOffset : 0; - const sliced = - limit !== undefined ? filtered.slice(offset, offset + limit) : filtered; - + const page = parsePage(req.query_params); return { status_code: 200, body: { - memories: sliced, + memories: takePage(filtered, page), total: filtered.length, - offset, - limit: limit ?? null, + ...page, }, }; }, @@ -2233,7 +2251,11 @@ export function registerApiTriggers( const authErr = checkAuth(req, secret); if (authErr) return authErr; const semantic = await kv.list(KV.semantic); - return { status_code: 200, body: { semantic } }; + const page = parsePage(req.query_params); + return { + status_code: 200, + body: { semantic: takePage(semantic, page), total: semantic.length, ...page }, + }; }, ); sdk.registerTrigger({ @@ -2247,7 +2269,11 @@ export function registerApiTriggers( const authErr = checkAuth(req, secret); if (authErr) return authErr; const procedural = await kv.list(KV.procedural); - return { status_code: 200, body: { procedural } }; + const page = parsePage(req.query_params); + return { + status_code: 200, + body: { procedural: takePage(procedural, page), total: procedural.length, ...page }, + }; }, ); sdk.registerTrigger({ diff --git a/src/viewer/index.html b/src/viewer/index.html index 75d9e2251e..c3a46d49f0 100644 --- a/src/viewer/index.html +++ b/src/viewer/index.html @@ -1473,7 +1473,7 @@

agentmemory

try { var results = await Promise.all([ api('health', { readErrorBody: true }), - apiGet('sessions'), + apiGet('sessions?limit=all'), apiGet('memories?latest=true&limit=500'), apiGet('graph/stats'), apiGet('audit?limit=5'), @@ -2801,7 +2801,7 @@

agentmemory

async function loadTimeline() { var el = document.getElementById('view-timeline'); if (!state.timeline.loaded) el.innerHTML = '
Loading timeline...
'; - var sessResult = await apiGet('sessions'); + var sessResult = await apiGet('sessions?limit=all'); var sessions = (sessResult && sessResult.sessions) || []; state.timeline.loaded = true; @@ -3034,7 +3034,7 @@

agentmemory

var el = document.getElementById('view-activity'); if (!state.activity.loaded) el.innerHTML = '
Loading activity...
'; var results = await Promise.all([ - apiGet('sessions'), + apiGet('sessions?limit=all'), apiGet('audit?limit=200') ]); var sessions = (results[0] && results[0].sessions) || []; @@ -3170,7 +3170,7 @@

agentmemory

async function loadSessions() { var el = document.getElementById('view-sessions'); if (!state.sessions.loaded) el.innerHTML = '
Loading sessions...
'; - var result = await apiGet('sessions'); + var result = await apiGet('sessions?limit=all'); state.sessions.items = (result && result.sessions) || []; state.sessions.loaded = true; renderSessions(); @@ -3682,7 +3682,7 @@

agentmemory

async function loadProfile() { var el = document.getElementById('view-profile'); el.innerHTML = '
Loading profile...
'; - var sessResult = await apiGet('sessions'); + var sessResult = await apiGet('sessions?limit=all'); var sessions = (sessResult && sessResult.sessions) || []; var projects = {}; diff --git a/test/fetch-timeout.test.ts b/test/fetch-timeout.test.ts index f4a4286780..11ea970261 100644 --- a/test/fetch-timeout.test.ts +++ b/test/fetch-timeout.test.ts @@ -79,7 +79,7 @@ describe("fetchWithTimeout", () => { // // The retry wrapper must bound TOTAL elapsed time across every // attempt + sleep — not per-attempt — so worst case never blows -// past the caller's budget or the iii 180s invocation timeout. +// past the caller's configured budget. // ───────────────────────────────────────────────────────────── describe("fetchWithTimeout bounded retry (total deadline)", () => { // Builds a fetch mock that replays a queue of {status, headers} in order, @@ -106,12 +106,12 @@ describe("fetchWithTimeout bounded retry (total deadline)", () => { delete process.env["AGENTMEMORY_LLM_TIMEOUT_MS"]; }); - // 0. The FIRST attempt must honor the hard budget cap, not the raw caller - // timeout. Regression: fetchOnce was called with `ms` instead of the - // capped `budgetMs`, so a large caller timeout (e.g. 300s) could hang the - // initial request past HARD_BUDGET_CAP_MS (170s) and the iii 180s - // invocation timeout before any retry logic ran. - it("caps the first attempt at the hard budget, not the raw caller timeout", async () => { + // 0. The FIRST attempt must honor the caller's configured timeout in full. + // Regression: a 170s HARD_BUDGET_CAP_MS silently shortened every budget + // above it, so AGENTMEMORY_LLM_TIMEOUT_MS=480000 aborted at 170s while the + // error message still reported 480000ms. Slow local models legitimately + // run past 170s; the configured bound is the only bound. + it("honors the full caller timeout on the first attempt", async () => { const signals: AbortSignal[] = []; const capturing = ((_url: string, init?: RequestInit) => { const signal = init!.signal as AbortSignal; @@ -125,15 +125,17 @@ describe("fetchWithTimeout bounded retry (total deadline)", () => { vi.spyOn(globalThis, "fetch").mockImplementation(capturing); vi.useFakeTimers(); - // 300s caller timeout — far above the 170s HARD_BUDGET_CAP_MS. + // 300s caller timeout — well past the 170s cap this used to be clamped to. const p = fetchWithTimeout("https://example.com", {}, 300000); p.catch(() => {}); // observe rejection so it isn't flagged unhandled expect(signals).toHaveLength(1); - // Just before the cap the first attempt is still alive. - await vi.advanceTimersByTimeAsync(169999); + // Past the old 170s cap the attempt must still be alive. + await vi.advanceTimersByTimeAsync(200000); + expect(signals[0].aborted).toBe(false); + // It aborts only at the configured 300s bound. + await vi.advanceTimersByTimeAsync(99999); expect(signals[0].aborted).toBe(false); - // At the cap it must abort. Pre-fix (raw 300s) it would still be alive. await vi.advanceTimersByTimeAsync(2); expect(signals[0].aborted).toBe(true); diff --git a/test/list-routes-paging.test.ts b/test/list-routes-paging.test.ts new file mode 100644 index 0000000000..8d68a59e74 --- /dev/null +++ b/test/list-routes-paging.test.ts @@ -0,0 +1,203 @@ +import { describe, it, expect, beforeEach, vi } from "vitest"; + +vi.mock("../src/logger.js", () => ({ + logger: { info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() }, +})); + +vi.mock("../src/state/keyed-mutex.js", () => ({ + withKeyedLock: (_key: string, fn: () => Promise) => fn(), +})); + +import { registerApiTriggers } from "../src/triggers/api.js"; +import { KV } from "../src/state/schema.js"; +import type { Session } from "../src/types.js"; + +const SECRET = "paging-test-secret"; + +function mockKV() { + const store = new Map>(); + return { + get: async (scope: string, key: string): Promise => + (store.get(scope)?.get(key) as T) ?? null, + set: async (scope: string, key: string, data: T): Promise => { + if (!store.has(scope)) store.set(scope, new Map()); + store.get(scope)!.set(key, data); + return data; + }, + delete: async (scope: string, key: string): Promise => { + store.get(scope)?.delete(key); + }, + list: async (scope: string): Promise => + Array.from(store.get(scope)?.values() ?? []) as T[], + }; +} + +function mockSdk() { + const fns = new Map(); + return { + registerFunction: (id: string, h: Function) => { + fns.set(id, h); + }, + registerTrigger: () => {}, + trigger: async (input: { function_id: string; payload?: unknown }) => + fns.get(input.function_id)?.(input.payload), + _fns: fns, + }; +} + +type Route = "api::sessions" | "api::semantic-list" | "api::procedural-list" | "api::memories"; + +let kv: ReturnType; +let sdk: ReturnType; + +async function get(route: Route, query: Record = {}) { + return sdk._fns.get(route)!({ + headers: { authorization: `Bearer ${SECRET}` }, + query_params: query, + }) as Promise<{ status_code: number; body: Record }>; +} + +function makeSession(n: number): Session { + return { + id: `ses_${n}`, + project: "proj", + cwd: "/proj", + startedAt: new Date(Date.UTC(2026, 0, 1, 0, n)).toISOString(), + status: "completed", + observationCount: 0, + }; +} + +async function seedSessions(count: number) { + const order = Array.from({ length: count }, (_, i) => i); + for (let i = order.length - 1; i > 0; i--) { + const j = (i * 7919) % (i + 1); + [order[i], order[j]] = [order[j]!, order[i]!]; + } + for (const n of order) await kv.set(KV.sessions, `ses_${n}`, makeSession(n)); +} + +async function seedRows(scope: string, count: number) { + for (let n = 0; n < count; n++) { + await kv.set(scope, `row_${n}`, { id: `row_${n}`, n }); + } +} + +beforeEach(() => { + kv = mockKV(); + sdk = mockSdk(); + registerApiTriggers(sdk as never, kv as never, SECRET); +}); + +describe("GET /agentmemory/sessions paging", () => { + it("returns the newest 100 by default with the full total", async () => { + await seedSessions(130); + const res = await get("api::sessions"); + const sessions = res.body.sessions as Session[]; + expect(sessions).toHaveLength(100); + expect(res.body.total).toBe(130); + expect(res.body.limit).toBe(100); + expect(res.body.offset).toBe(0); + expect(sessions[0]!.id).toBe("ses_129"); + expect(sessions[99]!.id).toBe("ses_30"); + }); + + it("orders newest-first before paging and honours limit/offset", async () => { + await seedSessions(25); + const res = await get("api::sessions", { limit: "10", offset: "10" }); + const ids = (res.body.sessions as Session[]).map((s) => s.id); + expect(ids).toEqual(Array.from({ length: 10 }, (_, i) => `ses_${14 - i}`)); + expect(res.body.total).toBe(25); + expect(res.body.limit).toBe(10); + expect(res.body.offset).toBe(10); + }); + + it("returns every Session for limit=all", async () => { + await seedSessions(130); + const res = await get("api::sessions", { limit: "all" }); + expect(res.body.sessions).toHaveLength(130); + expect(res.body.total).toBe(130); + expect(res.body.limit).toBe("all"); + }); + + it("sorts a Session row without startedAt last instead of throwing", async () => { + await seedSessions(3); + await kv.set(KV.sessions, "ses_ghost", { endedAt: "2026-09-01T00:00:00.000Z", status: "completed" }); + const res = await get("api::sessions"); + const ids = (res.body.sessions as Array<{ id?: string }>).map((s) => s.id); + expect(ids).toEqual(["ses_2", "ses_1", "ses_0", undefined]); + expect(res.body.total).toBe(4); + }); + + it("counts the total after the agent filter", async () => { + await seedSessions(5); + await kv.set(KV.sessions, "ses_other", { ...makeSession(99), id: "ses_other", agentId: "other" }); + const res = await get("api::sessions", { agentId: "other", limit: "2" }); + expect(res.body.sessions).toHaveLength(1); + expect(res.body.total).toBe(1); + }); + + it("falls back to the defaults for a malformed limit or offset", async () => { + await seedSessions(3); + const res = await get("api::sessions", { limit: "-4", offset: "x" }); + expect(res.body.sessions).toHaveLength(3); + expect(res.body.limit).toBe(100); + expect(res.body.offset).toBe(0); + }); + + it("still attaches the stored Session Summary to the page it returns", async () => { + await seedSessions(3); + await kv.set(KV.summaries, "ses_2", { sessionId: "ses_2", request: "r" }); + const res = await get("api::sessions", { limit: "1" }); + const [first] = res.body.sessions as Array; + expect(first!.id).toBe("ses_2"); + expect(first!.summary).toEqual({ sessionId: "ses_2", request: "r" }); + }); +}); + +describe.each([ + ["api::semantic-list" as const, KV.semantic, "semantic"], + ["api::procedural-list" as const, KV.procedural, "procedural"], + ["api::memories" as const, KV.memories, "memories"], +])("GET %s paging", (route, scope, key) => { + it("returns at most 100 rows plus the full total", async () => { + await seedRows(scope, 120); + const res = await get(route); + expect(res.body[key]).toHaveLength(100); + expect(res.body.total).toBe(120); + expect(res.body.limit).toBe(100); + expect(res.body.offset).toBe(0); + }); + + it("pages with limit/offset", async () => { + await seedRows(scope, 12); + const res = await get(route, { limit: "5", offset: "10" }); + expect(res.body[key]).toHaveLength(2); + expect(res.body.total).toBe(12); + expect(res.body.offset).toBe(10); + }); + + it("returns everything for limit=all", async () => { + await seedRows(scope, 120); + const res = await get(route, { limit: "all" }); + expect(res.body[key]).toHaveLength(120); + expect(res.body.limit).toBe("all"); + }); +}); + +describe("GET /agentmemory/memories paging", () => { + it("pages the latest-filtered set and counts total after the filter", async () => { + for (let n = 0; n < 6; n++) { + await kv.set(KV.memories, `mem_${n}`, { id: `mem_${n}`, isLatest: n % 2 === 0 }); + } + const res = await get("api::memories", { latest: "true", limit: "2", offset: "1" }); + expect((res.body.memories as Array<{ id: string }>).map((m) => m.id)).toEqual(["mem_2", "mem_4"]); + expect(res.body.total).toBe(3); + }); + + it("still answers count=true with totals only", async () => { + await seedRows(KV.memories, 3); + const res = await get("api::memories", { count: "true" }); + expect(res.body).toEqual({ total: 3, latestCount: 0 }); + }); +}); diff --git a/test/memories-pagination.test.ts b/test/memories-pagination.test.ts index c3275df617..f12c579a70 100644 --- a/test/memories-pagination.test.ts +++ b/test/memories-pagination.test.ts @@ -14,17 +14,6 @@ describe("memories + export pagination (#544)", () => { expect(api).toMatch(/latestCount:\s*filtered\.filter/); }); - it("api::memories accepts limit + offset query params", () => { - expect(api).toMatch(/query_params\?\.\["limit"\]/); - expect(api).toMatch(/query_params\?\.\["offset"\]/); - expect(api).toMatch(/filtered\.slice\(offset/); - expect(api).toMatch(/total:\s*filtered\.length/); - }); - - it("api::memories caps limit at 5000 to bound response size", () => { - expect(api).toMatch(/Math\.min\(parsedLimit,\s*5000\)/); - }); - it("api::export passes through maxSessions + offset query params", () => { expect(api).toMatch(/query_params\?\.\["maxSessions"\]/); expect(api).toMatch(/query_params\?\.\["offset"\]/); diff --git a/test/search.test.ts b/test/search.test.ts index 84b8134e3a..c277b0f45d 100644 --- a/test/search.test.ts +++ b/test/search.test.ts @@ -150,6 +150,63 @@ describe("mem::search", () => { expect(result.truncated).toBe(true); }); + it("skips an oversized record and keeps the smaller ones that fit under the budget", async () => { + const oversized: CompressedObservation = { + id: "obs_big", + sessionId: "ses_1", + timestamp: "2026-01-03T00:00:00Z", + type: "decision", + title: "Auth middleware decision revisited", + facts: ["Auth middleware keeps JWT"], + narrative: "auth middleware ".repeat(400), + concepts: ["auth", "jwt"], + files: ["src/auth.ts"], + importance: 9, + }; + const small: CompressedObservation = { + id: "obs_small", + sessionId: "ses_1", + timestamp: "2026-01-04T00:00:00Z", + type: "decision", + title: "Auth middleware follow-up", + facts: ["Auth middleware logs"], + narrative: "Added auth middleware logging.", + concepts: ["auth"], + files: ["src/auth.ts"], + importance: 8, + }; + await kv.set(KV.observations("ses_1"), oversized.id, oversized); + await kv.set(KV.observations("ses_1"), small.id, small); + getSearchIndex().clear(); + await rebuildIndex(kv as never); + + const result = (await sdk.trigger("mem::search", { + query: "auth middleware", + token_budget: 600, + })) as { + results: Array<{ observation: CompressedObservation }>; + truncated: boolean; + tokens_used: number; + }; + + const ids = result.results.map((r) => r.observation.id); + expect(ids).not.toContain("obs_big"); + expect(ids).toEqual(expect.arrayContaining(["obs_a", "obs_small"])); + expect(result.tokens_used).toBeLessThanOrEqual(600); + expect(result.truncated).toBe(true); + }); + + it("reports truncated only when a record was skipped", async () => { + const result = (await sdk.trigger("mem::search", { + query: "auth middleware", + format: "compact", + token_budget: 10_000, + })) as { results: Array<{ obsId: string }>; truncated: boolean }; + + expect(result.results.map((r) => r.obsId)).toEqual(["obs_a"]); + expect(result.truncated).toBe(false); + }); + it("rejects invalid format values", async () => { await expect( sdk.trigger("mem::search", { query: "auth", format: "verbose" }), diff --git a/test/session-end-unknown.test.ts b/test/session-end-unknown.test.ts new file mode 100644 index 0000000000..1aa3e561e1 --- /dev/null +++ b/test/session-end-unknown.test.ts @@ -0,0 +1,115 @@ +import { describe, it, expect, beforeEach, vi } from "vitest"; + +vi.mock("../src/logger.js", () => ({ + logger: { info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() }, +})); + +vi.mock("../src/state/keyed-mutex.js", () => ({ + withKeyedLock: (_key: string, fn: () => Promise) => fn(), +})); + +import { registerApiTriggers } from "../src/triggers/api.js"; +import { KV } from "../src/state/schema.js"; +import type { Session } from "../src/types.js"; + +const SECRET = "session-end-test-secret"; + +function mockKV() { + const store = new Map>(); + const scope = (name: string) => { + if (!store.has(name)) store.set(name, new Map()); + return store.get(name)!; + }; + return { + get: async (s: string, key: string): Promise => + (store.get(s)?.get(key) as T) ?? null, + set: async (s: string, key: string, data: T): Promise => { + scope(s).set(key, data); + return data; + }, + delete: async (s: string, key: string): Promise => { + store.get(s)?.delete(key); + }, + update: async ( + s: string, + key: string, + ops: Array<{ type: string; path: string; value: unknown }>, + ) => { + const base = { ...((store.get(s)?.get(key) as object) ?? {}) } as Record; + for (const op of ops) base[op.path] = op.value; + scope(s).set(key, base); + }, + list: async (s: string): Promise => + Array.from(store.get(s)?.values() ?? []) as T[], + }; +} + +function mockSdk() { + const fns = new Map(); + return { + registerFunction: (id: string, h: Function) => { + fns.set(id, h); + }, + registerTrigger: () => {}, + trigger: async (input: { function_id: string; payload?: unknown }) => + fns.get(input.function_id)?.(input.payload), + _fns: fns, + }; +} + +let kv: ReturnType; +let sdk: ReturnType; +let stopped: string[]; + +function endSession(body: Record) { + return sdk._fns.get("api::session::end")!({ + headers: { authorization: `Bearer ${SECRET}` }, + body, + }) as Promise<{ status_code: number; body: Record }>; +} + +beforeEach(() => { + kv = mockKV(); + sdk = mockSdk(); + stopped = []; + registerApiTriggers(sdk as never, kv as never, SECRET); + sdk.registerFunction("event::session::stopped", async (data: { sessionId: string }) => { + stopped.push(data.sessionId); + }); +}); + +describe("POST /agentmemory/session/end", () => { + it("answers 404 session_not_found for a Session that never started and writes nothing", async () => { + const res = await endSession({ sessionId: "ses_never" }); + + expect(res.status_code).toBe(404); + expect(res.body).toEqual({ error: "session_not_found" }); + expect(await kv.list(KV.sessions)).toEqual([]); + expect(stopped).toEqual([]); + }); + + it("ends a known Session as before", async () => { + await kv.set(KV.sessions, "ses_1", { + id: "ses_1", + project: "proj", + cwd: "/proj", + startedAt: "2026-09-01T00:00:00.000Z", + status: "active", + observationCount: 2, + } satisfies Session); + + const res = await endSession({ sessionId: "ses_1" }); + + expect(res.status_code).toBe(200); + expect(res.body).toEqual({ success: true }); + const session = await kv.get(KV.sessions, "ses_1"); + expect(session?.status).toBe("completed"); + expect(typeof session?.endedAt).toBe("string"); + expect(stopped).toEqual(["ses_1"]); + }); + + it("still rejects a missing sessionId with 400", async () => { + const res = await endSession({}); + expect(res.status_code).toBe(400); + }); +});