vantage-peers-mcp 2.10.0 → 2.12.0

package/CHANGELOG.md

@@ -1,5 +1,69 @@
 # Changelog
 
+## [2.12.0] — 2026-06-14
+
+### Changed
+- `check_messages` now calls `messages:checkNewMessagesEnvelope` (Convex PR #759 mergeCommit `c03023fc85463fe16fc53948d3002420bc1c4547`) for bounded tool-response size. New optional `limit` arg (1-50, default 20). When the call is truncated, the text payload appends `— truncated. Resume with check_messages since=<nextSince>` so the orchestrator can page the backlog cleanly without busting Claude Code's tool-response cap.
+- `since` arg description amended with a "pair with `nextSince` from a previous truncated reply" hint.
+
+### Notes
+- Closes Pi BLOCKER VP task `k1702xaahb` (Day 102 — sigma-vps cron crash on a 36-message / 53 KB backlog).
+- Legacy `messages:checkNewMessages` Convex query left intact for vp-mcp <2.12.0 callers (no break).
+
+## [2.11.0] — 2026-06-14 — Day 102 CRUD baseline PR-C-bis option B: 3-entity Convex searchIndex (mission k575kc1ryps0n8br95jw3q7d0x88m2v9)
+
+Mission `mcp-crud-baseline-standard` PR-C-bis under T2. Pi-sequenced follow-up after PR-C (rename-only safe subset) — implements **option B SCOPED** per Pi arbitrage msg `jn7abynmghy5qdr9ga0b914wmh88n99w` ("GO option B SCOPED — démarre PR-C avec 3 entités prioritaires"): tasks + messages + briefingNotes get native Convex BM25 search via `.searchIndex()` schema additions + per-entity Convex queries + MCP tool wrappers.
+
+Backend choice (B over A): Convex native `.searchIndex()` is BM25-only, no embeddings, no RAG per-entity pipeline. Sub-linear scan stays in the same backend without spinning up an embedding pipeline per table. Hybrid/semantic per entity remains a future RFC.
+
+### Added — Convex schema searchIndex (3 tables)
+
+- `tasks` → `searchIndex("search_title", { searchField: "title", filterFields: ["assignedTo", "status", "project", "missionId", "orgId"] })`
+- `messages` → `searchIndex("search_content", { searchField: "content", filterFields: ["from", "channel", "sessionDay", "tenantId"] })`
+- `briefingNotes` → `searchIndex("search_content", { searchField: "content", filterFields: ["topic", "createdBy", "orgId"] })`
+
+### Added — Convex query functions
+
+- `tasks:searchTasksByKeyword` — BM25 over title with assignedTo/status/project/missionId filters + scope gate via `withOrgScope` + `filterByOrgScope` + `requireScope("view-own-tasks")`. Lite/full projection.
+- `messages:searchMessagesByKeyword` — BM25 over content with from/channel/sessionDay/tenantId filters.
+- `briefingNotes:searchBriefingNotesByKeyword` — BM25 over content with topic/createdBy filters.
+
+All three: 20-item default limit, 200 cap, lite projection on demand.
+
+### Added — MCP tool wrappers (3 canonical)
+
+- **`search_tasks_by_keyword`** — wires to `tasks:searchTasksByKeyword`. `readOnlyHint=true`.
+- **`search_messages_by_keyword`** — wires to `messages:searchMessagesByKeyword`. `readOnlyHint=true`.
+- **`search_briefing_notes_by_keyword`** — wires to `briefingNotes:searchBriefingNotesByKeyword`. `readOnlyHint=true`.
+
+### Why option B (and why scoped to 3)
+
+`grep searchIndex convex/schema.ts` previously returned zero hits; only `memories` had search infrastructure (via `@convex-dev/rag`). The original "13-entity cluster" full scope was deferred via PR-C (rename-only) after arbitrage timeout. Pi's arbitrage landed shortly after: **option B SCOPED to 3 priority entities** is the right blend of doctrine progress + bounded scope:
+- `tasks` — largest fleet audience (Eta T13 close-issue scan, dispatch-task-find, etc.).
+- `messages` — post-incident audit + messages-history skill demand.
+- `briefingNotes` — daily snapshot recall narrative.
+
+The remaining 10 entities (mission/mandate/fix_pattern/component/repo_mapping/bu/profile/deployment/diary/error/issue/recurring_task/summary) go to a per-entity RFC (mission walk-through scheduled Wed 17 June 15h with Laurent) — each entity decides B (searchIndex), A (RAG if semantic justified, e.g. episodes/diary), or N/A semantic documented (deployment/repo_mapping).
+
+### Version sync
+
+- `mcp-server/server.ts:115` SERVER_VERSION 2.10.0 → 2.11.0
+- `mcp-server/package.json` → 2.11.0
+- README + 4 cloud docs bumped to 2.11.0 markers (`enforce-release-sync` v1.0.1 gate).
+
+### Test fixture catch-up
+
+- `READ_ONLY_TOOLS` set in `mcp-server/src/__tests__/chatgpt-tool-annotations.test.ts` extended with the 3 new canonical tool names.
+
+### Refs
+
+- Mission `k575kc1ryps0n8br95jw3q7d0x88m2v9`.
+- Pi sequencing dispatch: msg `jn76360ckrrkwpqbfwa6tst7k588mpsh` ("(1) T2 PR-C-bis option B SCOPED 3 entities").
+- Pi arbitrage that selected option B: msg `jn7abynmghy5qdr9ga0b914wmh88n99w`.
+- Audit T1: `analysis/mcp-crud-baseline-vp-audit-2026-06-14.md` rows 1 (task), 3 (message), 4 (briefing_note).
+- Doctrine memory: `j57dhrmkzjerjtssnr0z9ba57n88n7q7` (5 ops per entity).
+- Predecessors: PR-A 2.8.0 (memories canonical), PR-B 2.9.0 (episode 5-op), PR-C 2.10.0 (rename-only safe subset).
+
 ## [2.10.0] — 2026-06-14 — Day 102 CRUD baseline PR-C: rename-only safe subset (mission k575kc1ryps0n8br95jw3q7d0x88m2v9)
 
 Mission `mcp-crud-baseline-standard` PR-C under T2. Third of 4 sub-PRs aligning the MCP surface with the Day 101 doctrine `j57dhrmkzjerjtssnr0z9ba57n88n7q7`. Sigma autonomous default after arbitrage timeout on the original "13-entity search_by_keyword cluster" scope — that fuller scope requires a backend search-infrastructure decision (RAG-index per entity vs Convex `.searchIndex()` per table) that is NOT a thin-wrapper PR. PR-C ships the rename-only safe subset now to keep doctrine velocity; the full cluster is deferred to a backend-RFC mission.

package/README.md

@@ -91,7 +91,7 @@ The server also reads `CONVEX_URL` from `.env.local` in the parent directory if
 ## Tools (97)
 
 ### Memory (6)
-`store_memory`, `recall`, `list_memories`, `soft_delete_memory`, `get_memory`, `store_episode`
+`store_memory`, `search_memories_by_semantic` (alias `recall`), `list_memories`, `soft_delete_memory`, `get_memory`, `store_episode`
 
 ### Profiles (3)
 `get_profile`, `update_profile`, `list_peers`
@@ -115,7 +115,7 @@ The server also reads `CONVEX_URL` from `.env.local` in the parent directory if
 `create_briefing_note`, `list_briefing_notes`
 
 ### Search / RAG (3)
-`search_fix_patterns`, `text_search`, `hybrid_search`
+`search_fix_patterns_by_semantic` (alias `search_fix_patterns`), `search_memories_by_keyword` (alias `text_search`), `hybrid_search`
 
 ### Issues (6)
 `get_issue`, `list_issues`, `update_issue_status`, `verify_issue`, `issue_stats`, `link_commit_to_issue`
@@ -231,7 +231,7 @@ Example:
 `create_bu`, `list_bus`, `get_bu`, `update_bu`, `delete_bu`
 
 ### Components (6)
-`register_component`, `list_components`, `get_component`, `update_component`, `delete_component`, `search_components`
+`register_component`, `list_components`, `get_component`, `update_component`, `delete_component`, `search_components_by_keyword` (alias `search_components`)
 
 ### Mandates (6)
 `create_mandate`, `list_mandates`, `accept_mandate`, `update_mandate`, `validate_mandate_spending`, `settle_mandate`
@@ -242,7 +242,7 @@ Example:
 ### Session (1)
 `set_summary`
 
-## Compact payloads and status aliases (v2.10.0 — feature since v2.3.0)
+## Compact payloads and status aliases (v2.11.0 — feature since v2.3.0)
 
 ### `fields=lite` — reduced token payloads
 
@@ -312,13 +312,13 @@ A fix pattern is a validated learning extracted from a resolved bug — symptom,
 
 The cycle runs as follows:
 
-1. **Agent encounters a bug.** Before touching any code, call `search_fix_patterns` with a plain-language description of the symptom. The KB returns ranked matches using semantic vector search.
+1. **Agent encounters a bug.** Before touching any code, call `search_fix_patterns_by_semantic` (alias `search_fix_patterns`) with a plain-language description of the symptom. The KB returns ranked matches using semantic vector search.
 2. **KB hit.** If a validated pattern is returned, apply the known fix directly. Log the reuse via `add_fix_attempt` (`worked: true`) so confidence scores stay current.
 3. **KB miss.** If no pattern matches, the agent fixes the bug manually using standard debugging. Once resolved, the learning is captured immediately via `create_fix_pattern` — symptom, root cause, severity, stack, and the working fix.
 4. **Validation.** After the fix holds in production (or after a second independent confirmation), call `validate_fix` to promote the pattern to validated status. This is the signal that downstream agents can trust the pattern without verification.
 5. **Issue linkage.** Call `link_issue_to_pattern` to attach the VantagePeers issue ID to the pattern. This creates a bidirectional reference: the issue record points to the pattern, and the pattern's `linkedIssueIds` list points back.
 
-The four tools that power this cycle are: `create_fix_pattern`, `add_fix_attempt`, `validate_fix`, and `link_issue_to_pattern`. The fifth tool, `search_fix_patterns`, is in the Search / RAG category and is the entry point agents should call first.
+The four tools that power this cycle are: `create_fix_pattern`, `add_fix_attempt`, `validate_fix`, and `link_issue_to_pattern`. The fifth tool, `search_fix_patterns_by_semantic` (alias `search_fix_patterns`), is in the Search / RAG category and is the entry point agents should call first.
 
 On the agent side, the `/capitalize-fix` skill and the `inject-fix-patterns` hook automate steps 3-5: the hook fires on task completion events and prompts the orchestrator to capture the learning before closing the task. The cycle is designed to be low-friction — one tool call per step, all via MCP, no `npx convex run` required.
 

package/dist/server.d.ts

@@ -1,12 +1,19 @@
 #!/usr/bin/env node
 /**
- * VantagePeers MCP Server
- * Exposes 82 Convex-backed tools to Claude Code agents via stdio transport.
+ * VantagePeers MCP Server — stdio transport (Self-host / local Claude Code path).
  *
- * Tool categories: Memory, Profiles, Messages, Tasks, Missions, Diary,
- * Briefing Notes, Components, Recurring Tasks, Mandates, Business Units,
- * Issues, Fix Patterns, Search/RAG, Mission Templates, Error Monitoring,
- * Deployments, Repo Mappings.
+ * Thin bootstrap: resolves CONVEX_URL, instantiates McpServer + ConvexHttpClient,
+ * delegates ALL tool registration to the shared `registerTools(server, convex)`
+ * surface in src/tools.ts. This guarantees stdio and HTTP transports expose the
+ * same tool set (parity locked by src/__tests__/stdio-http-parity.test.ts).
+ *
+ * Day 102 refactor (fix-pattern m974adhs7wtwb6pt4h0sdxn1k988mnxk):
+ * Removed ~3.6k LOC of inline `server.tool(...)` registrations that had drifted
+ * out of sync with src/tools.ts, hiding 12 new CRUD-baseline tools from fleet
+ * stdio consumers (Claude Code, Codex via npx vantage-peers-mcp).
+ *
+ * VP_EMIT_UI_MARKERS env gate is honored inside src/tools.ts and therefore
+ * works transparently on stdio as well as HTTP.
  *
  * See README.md for the full tool reference.
  */