agent-working-memory 0.7.3 → 0.7.6

package/README.md

@@ -407,7 +407,30 @@ npm run test:locomo   # LoCoMo industry benchmark (28.2%)
 
 All three ML models run locally via ONNX. No external API calls for retrieval. The entire system is a single SQLite file + a Node.js process.
 
-## What's New in v0.7.1
+## What's New in v0.7.6
+
+- **Recall latency 11-23s → 2.5s end-to-end (~5× faster)** — measurement spike found the slow path was a SQLite query-plan trap, not vector search. The BM25 query `JOIN engrams_fts ON e.rowid + WHERE MATCH + ORDER BY rank LIMIT N` materialized all matching rows (with 1.5KB embedding blobs) before the LIMIT applied. CTE prefilter forces FTS5 LIMIT first, then joins only the top-K rowids. Same SQLite, same data, same results — 567× faster for wide OR queries (3682ms → 6.5ms verified). Also added `getAssociationsForBatch` to replace the per-candidate N+1 in the activation scoring loop. Top-K results are byte-identical to the old query (verified by the equivalence test in `spike/`).
+- **Salience filter — auto-promote verified operational records** — operational batch summaries (e.g., "Submitted 6 events 2026-05-07 — IDs 18969, 18971…") were being discarded at salience 0.14 because BM25 novelty couldn't distinguish "useful new operational record" from "duplicate observation" when topic terminology repeated. New `detectVerifiedFinding()` pattern detector parallel to `detectUserFeedback()`: requires action-verb header (Submitted/Finalized/Completed/Reconciled/Triaged/etc.) plus ≥2 concrete identifiers (ISO date or contextual numeric ID). Matched memories get a 0.45 salience floor (active disposition, not canonical). 7 new tests, 23 salience tests pass.
+
+## What's New in v0.7.4
+
+- **Channel push telemetry** — new `GET /telemetry/channels` JSON endpoint and Prometheus counters (`coord_channel_push_attempts_total`, `..._delivered_total`, `..._failed_total{reason}`, `..._no_session_total`, `..._fallback_mailbox_total`, `..._session_disconnects_total`). Surfaces real delivery rate so coordination reliability can be measured rather than guessed.
+- **Role-based `/channel/push` addressing** — accepts `{role, workspace, message}` as alternative to `{agentId, message}`. Server resolves role+workspace to most-recently-seen alive agent. Lets workers notify the coordinator without hardcoding its UUID (which changes across coordinator restarts). Enables event-driven worker → coordinator hand-off in place of fragile coordinator self-polling.
+- **`/checkin` writes role on every call** — previously the UPDATE on existing rows preserved a stale role from initial registration; now agents can correct their own role via re-checkin.
+- **`/workers` JOINs channel sessions** — `alive` field is now `recent_pulse OR connected_channel_session`. Stops false-dead duplicate-spawn loops where a busy worker's `/pulse` went stale during long tool sequences while their channel-server stayed reachable.
+- **`cleanupStale` runs on a 5-minute schedule** — was only invoked manually; now zombie agents get marked dead automatically with a 600s threshold (forgiving for long edits).
+- **`user_feedback` salience event type** — new event type with bonus 0.3 (highest of any). Auto-detect heuristic on `memory_write` content matching `^(Robert|Katherine|Nancy|...) (said|verbatim|directed|decided|...)` forces `memoryClass='canonical'` so user-stated decisions can't be discarded by the BM25 novelty floor in populated DBs.
+
+### v0.7.3
+
+- **Salience filter production tuning** — fixed BM25 novelty floor that was discarding ~17% salience for most writes in 10K+ engram DBs. Quadratic dampening curve (`max(0.05, 1 - topScore²)`); concept-match penalty scoped to last 30 days; floor lowered 0.10 → 0.05.
+- Maintenance scripts for backup pruning + lme/bench database cleanup.
+
+### v0.7.2
+
+- Workspace recall fix (was returning UUIDs not names in v0.7.1 release).
+
+### v0.7.1
 
 - **Agent-provided metadata tags** — `memory_write` accepts `project`, `topic`, `source`, `confidence_level`, `session_id`, `intent`. Stored as searchable prefixed tags (`proj=X`, `sid=Z`). Session ID tags alone improved LongMemEval recall 3x.
 - **Dual synthesis** — consolidation creates two types of summary memories: session summaries (tag-based, for perfect recall) and pattern syntheses (cross-session, for novel recall/creative connections).
@@ -434,7 +457,7 @@ See [CHANGELOG.md](CHANGELOG.md) for full details.
 
 ## Project Status
 
-AWM is in active development (v0.7.0). The core memory pipeline, consolidation system, multi-agent coordination, and MCP integration are stable and used daily in production coding workflows.
+AWM is in active development (v0.7.6). The core memory pipeline, consolidation system, multi-agent coordination, and MCP integration are stable and used daily in production coding workflows.
 
 - Core retrieval and consolidation: **stable**
 - MCP tools and Claude Code integration: **stable**

package/dist/api/routes.js

@@ -527,7 +527,7 @@ export function registerRoutes(app, deps) {
         const base = {
             status: 'ok',
             timestamp: new Date().toISOString(),
-            version: '0.7.2',
+            version: '0.7.6',
             coordination: coordEnabled,
         };
         if (coordEnabled) {

package/dist/cli.js

@@ -307,7 +307,7 @@ async function exportMemories() {
     // Collect unique agents
     const agents = [...new Set(memories.map((m) => m.agent_id))];
     const exportData = {
-        version: '0.7.2',
+        version: '0.7.6',
         exported_at: new Date().toISOString(),
         source_db: dbPath,
         agent_filter: agentFilter,

package/dist/coordination/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/coordination/index.ts"],"names":[],"mappings":"AAEA;;;GAGG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAMxD,OAAO,EAAkB,KAAK,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAGxE,mBAAmB,YAAY,CAAC;AAChC,OAAO,EAAE,KAAK,oBAAoB,EAAE,KAAK,kBAAkB,EAAE,MAAM,aAAa,CAAC;AACjF,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAQ/D,yEAAyE;AACzE,wBAAgB,WAAW,IAAI,oBAAoB,GAAG,IAAI,CAEzD;AAED,iEAAiE;AACjE,wBAAgB,qBAAqB,IAAI,OAAO,CAG/C;AAED,mGAAmG;AACnG,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,GAAG,IAAI,CAkGvG;AAED,qFAAqF;AACrF,wBAAsB,uBAAuB,IAAI,OAAO,CAAC,IAAI,CAAC,CAI7D"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/coordination/index.ts"],"names":[],"mappings":"AAEA;;;GAGG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAMxD,OAAO,EAAkB,KAAK,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAGxE,mBAAmB,YAAY,CAAC;AAChC,OAAO,EAAE,KAAK,oBAAoB,EAAE,KAAK,kBAAkB,EAAE,MAAM,aAAa,CAAC;AACjF,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAQ/D,yEAAyE;AACzE,wBAAgB,WAAW,IAAI,oBAAoB,GAAG,IAAI,CAEzD;AAED,iEAAiE;AACjE,wBAAgB,qBAAqB,IAAI,OAAO,CAG/C;AAED,mGAAmG;AACnG,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,GAAG,IAAI,CAqHvG;AAED,qFAAqF;AACrF,wBAAsB,uBAAuB,IAAI,OAAO,CAAC,IAAI,CAAC,CAI7D"}

package/dist/coordination/index.js

@@ -7,7 +7,7 @@
 import { ZodError } from 'zod';
 import { initCoordinationTables } from './schema.js';
 import { registerCoordinationRoutes } from './routes.js';
-import { cleanSlate, pruneOldHeartbeats, purgeDeadAgents } from './stale.js';
+import { cleanSlate, pruneOldHeartbeats, purgeDeadAgents, cleanupStale } from './stale.js';
 import { createWriteMutex, needsWriteLock } from './write-mutex.js';
 import { createEventBus } from './events.js';
 import { loadPlugins, teardownPlugins } from './plugin-loader.js';
@@ -87,6 +87,23 @@ export function initCoordination(app, db, store) {
         }
         catch { /* db may be closed */ }
     }, 60 * 60 * 1000));
+    // Periodic stale-agent cleanup every 5 min with 600s threshold (10 min idle).
+    // Forgiving for long-running edits — workers should pulse every 60s during active
+    // work, so 10 min without a pulse is genuinely dead. This catches the
+    // "alive but not seeing each other" pattern where workers' processes persist
+    // but their heartbeats stop. Without this scheduled, only an explicit
+    // POST /stale/cleanup call (made by the coordinator agent on startup) ever
+    // fires cleanupStale, leaving zombie agents accumulating between coordinator
+    // sessions.
+    cleanupIntervals.push(setInterval(() => {
+        try {
+            const result = cleanupStale(db, 600);
+            if (result.cleaned > 0) {
+                console.log(`  [stale-cleanup] auto-cleaned ${result.stale.length} stale agent(s), ${result.cleaned} resource(s) released`);
+            }
+        }
+        catch { /* db may be closed */ }
+    }, 5 * 60 * 1000));
     // Periodic channel liveness probe every 60s — mark unreachable sessions as disconnected
     cleanupIntervals.push(setInterval(async () => {
         try {

package/dist/coordination/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/coordination/index.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;GAGG;AAKH,OAAO,EAAE,QAAQ,EAAE,MAAM,KAAK,CAAC;AAC/B,OAAO,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAC;AACrD,OAAO,EAAE,0BAA0B,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,UAAU,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7E,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACpE,OAAO,EAAE,cAAc,EAA6B,MAAM,aAAa,CAAC;AACxE,OAAO,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAMlE,sDAAsD;AACtD,MAAM,gBAAgB,GAAqB,EAAE,CAAC;AAE9C,iEAAiE;AACjE,IAAI,oBAAoB,GAAgC,IAAI,CAAC;AAE7D,yEAAyE;AACzE,MAAM,UAAU,WAAW;IACzB,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,iEAAiE;AACjE,MAAM,UAAU,qBAAqB;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC;IACzC,OAAO,GAAG,KAAK,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC;AACvC,CAAC;AAED,mGAAmG;AACnG,MAAM,UAAU,gBAAgB,CAAC,GAAoB,EAAE,EAAqB,EAAE,KAAmB;IAC/F,0CAA0C;IAC1C,sBAAsB,CAAC,EAAE,CAAC,CAAC;IAE3B,gEAAgE;IAChE,UAAU,CAAC,EAAE,CAAC,CAAC;IAEf,mEAAmE;IACnE,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;QAChD,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,IAAI,EAAE,CAAC;QAC5C,IAAI,+BAA+B,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,kCAAkC,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;YACpG,KAAK,CAAC,MAAM,CAAC,6BAA6B,EAAE,MAAM,CAAC,CAAC;YACpD,KAAK,CAAC,MAAM,CAAC,8BAA8B,EAAE,wCAAwC,CAAC,CAAC;YACvF,KAAK,CAAC,MAAM,CAAC,8BAA8B,EAAE,6BAA6B,CAAC,CAAC;QAC9E,CAAC;QACD,IAAI,OAAO,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YACjC,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;QAChC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,0FAA0F;IAC1F,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC,YAAY,EAAE,EAAE;QACtC,IAAI,CAAC,YAAY,CAAC,SAAS,EAAE,CAAC;YAC5B,YAAY,CAAC,SAAS,GAAG,OAAO,CAAC;QACnC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,uEAAuE;IACvE,0DAA0D;IAC1D,MAAM,UAAU,GAAG,gBAAgB,EAAE,CAAC;IACtC,GAAG,CAAC,OAAO,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;QACjD,IAAI,cAAc,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YAChD,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,OAAO,EAAE,CAAC;YAC3C,gEAAgE;YAChE,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAClC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,8CAA8C;IAC9C,oBAAoB,GAAG,cAAc,EAAE,CAAC;IAExC,qCAAqC;IACrC,0BAA0B,CAAC,GAAG,EAAE,EAAE,EAAE,KAAK,EAAE,oBAAoB,CAAC,CAAC;IAEjE,qFAAqF;IACrF,GAAG,CAAC,eAAe,CAAC,CAAC,KAAsC,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE;QAC9E,IAAI,KAAK,YAAY,QAAQ,EAAE,CAAC;YAC9B,OAAO,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;gBAC5B,IAAI,EAAE,IAAI;gBACV,KAAK,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE;aAC5F,CAAC,CAAC;QACL,CAAC;QACD,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC;YACzC,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE;SAC1D,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,sFAAsF;IACtF,gBAAgB,CAAC,IAAI,CACnB,WAAW,CAAC,GAAG,EAAE;QACf,IAAI,CAAC;YAAC,kBAAkB,CAAC,EAAE,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IAClE,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,EAClB,WAAW,CAAC,GAAG,EAAE;QACf,IAAI,CAAC;YAAC,eAAe,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IACnE,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CACnB,CAAC;IAEF,wFAAwF;IACxF,gBAAgB,CAAC,IAAI,CACnB,WAAW,CAAC,KAAK,IAAI,EAAE;QACrB,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,EAAE,CAAC,OAAO,CACzB,oFAAoF,CACrF,CAAC,GAAG,EAAqD,CAAC;YAE3D,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;gBAC/B,IAAI,CAAC;oBACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,CAAC,UAAU,SAAS,EAAE;wBACtD,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC;qBAClC,CAAC,CAAC;oBACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;wBACZ,EAAE,CAAC,OAAO,CAAC,8EAA8E,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;oBACnH,CAAC;gBACH,CAAC;gBAAC,MAAM,CAAC;oBACP,EAAE,CAAC,OAAO,CAAC,8EAA8E,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;gBACnH,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IACpC,CAAC,EAAE,MAAM,CAAC,CACX,CAAC;IAEF,mEAAmE;IACnE,WAAW,CAAC,EAAE,MAAM,EAAE,oBAAoB,EAAE,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QAC5E,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAG,GAAa,CAAC,OAAO,CAAC,CAAC;IAC3E,CAAC,CAAC,CAAC;IAEH,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;AAC/C,CAAC;AAED,qFAAqF;AACrF,MAAM,CAAC,KAAK,UAAU,uBAAuB;IAC3C,KAAK,MAAM,EAAE,IAAI,gBAAgB;QAAE,aAAa,CAAC,EAAE,CAAC,CAAC;IACrD,gBAAgB,CAAC,MAAM,GAAG,CAAC,CAAC;IAC5B,MAAM,eAAe,EAAE,CAAC;AAC1B,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/coordination/index.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;GAGG;AAKH,OAAO,EAAE,QAAQ,EAAE,MAAM,KAAK,CAAC;AAC/B,OAAO,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAC;AACrD,OAAO,EAAE,0BAA0B,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,UAAU,EAAE,kBAAkB,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC3F,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACpE,OAAO,EAAE,cAAc,EAA6B,MAAM,aAAa,CAAC;AACxE,OAAO,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAMlE,sDAAsD;AACtD,MAAM,gBAAgB,GAAqB,EAAE,CAAC;AAE9C,iEAAiE;AACjE,IAAI,oBAAoB,GAAgC,IAAI,CAAC;AAE7D,yEAAyE;AACzE,MAAM,UAAU,WAAW;IACzB,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,iEAAiE;AACjE,MAAM,UAAU,qBAAqB;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC;IACzC,OAAO,GAAG,KAAK,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC;AACvC,CAAC;AAED,mGAAmG;AACnG,MAAM,UAAU,gBAAgB,CAAC,GAAoB,EAAE,EAAqB,EAAE,KAAmB;IAC/F,0CAA0C;IAC1C,sBAAsB,CAAC,EAAE,CAAC,CAAC;IAE3B,gEAAgE;IAChE,UAAU,CAAC,EAAE,CAAC,CAAC;IAEf,mEAAmE;IACnE,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;QAChD,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,IAAI,EAAE,CAAC;QAC5C,IAAI,+BAA+B,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,kCAAkC,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;YACpG,KAAK,CAAC,MAAM,CAAC,6BAA6B,EAAE,MAAM,CAAC,CAAC;YACpD,KAAK,CAAC,MAAM,CAAC,8BAA8B,EAAE,wCAAwC,CAAC,CAAC;YACvF,KAAK,CAAC,MAAM,CAAC,8BAA8B,EAAE,6BAA6B,CAAC,CAAC;QAC9E,CAAC;QACD,IAAI,OAAO,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YACjC,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;QAChC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,0FAA0F;IAC1F,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC,YAAY,EAAE,EAAE;QACtC,IAAI,CAAC,YAAY,CAAC,SAAS,EAAE,CAAC;YAC5B,YAAY,CAAC,SAAS,GAAG,OAAO,CAAC;QACnC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,uEAAuE;IACvE,0DAA0D;IAC1D,MAAM,UAAU,GAAG,gBAAgB,EAAE,CAAC;IACtC,GAAG,CAAC,OAAO,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;QACjD,IAAI,cAAc,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YAChD,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,OAAO,EAAE,CAAC;YAC3C,gEAAgE;YAChE,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAClC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,8CAA8C;IAC9C,oBAAoB,GAAG,cAAc,EAAE,CAAC;IAExC,qCAAqC;IACrC,0BAA0B,CAAC,GAAG,EAAE,EAAE,EAAE,KAAK,EAAE,oBAAoB,CAAC,CAAC;IAEjE,qFAAqF;IACrF,GAAG,CAAC,eAAe,CAAC,CAAC,KAAsC,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE;QAC9E,IAAI,KAAK,YAAY,QAAQ,EAAE,CAAC;YAC9B,OAAO,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;gBAC5B,IAAI,EAAE,IAAI;gBACV,KAAK,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE;aAC5F,CAAC,CAAC;QACL,CAAC;QACD,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC;YACzC,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE;SAC1D,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,sFAAsF;IACtF,gBAAgB,CAAC,IAAI,CACnB,WAAW,CAAC,GAAG,EAAE;QACf,IAAI,CAAC;YAAC,kBAAkB,CAAC,EAAE,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IAClE,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,EAClB,WAAW,CAAC,GAAG,EAAE;QACf,IAAI,CAAC;YAAC,eAAe,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IACnE,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CACnB,CAAC;IAEF,8EAA8E;IAC9E,kFAAkF;IAClF,sEAAsE;IACtE,6EAA6E;IAC7E,sEAAsE;IACtE,2EAA2E;IAC3E,6EAA6E;IAC7E,YAAY;IACZ,gBAAgB,CAAC,IAAI,CACnB,WAAW,CAAC,GAAG,EAAE;QACf,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,YAAY,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;YACrC,IAAI,MAAM,CAAC,OAAO,GAAG,CAAC,EAAE,CAAC;gBACvB,OAAO,CAAC,GAAG,CAAC,kCAAkC,MAAM,CAAC,KAAK,CAAC,MAAM,oBAAoB,MAAM,CAAC,OAAO,uBAAuB,CAAC,CAAC;YAC9H,CAAC;QACH,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IACpC,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,CAClB,CAAC;IAEF,wFAAwF;IACxF,gBAAgB,CAAC,IAAI,CACnB,WAAW,CAAC,KAAK,IAAI,EAAE;QACrB,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,EAAE,CAAC,OAAO,CACzB,oFAAoF,CACrF,CAAC,GAAG,EAAqD,CAAC;YAE3D,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;gBAC/B,IAAI,CAAC;oBACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,CAAC,UAAU,SAAS,EAAE;wBACtD,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC;qBAClC,CAAC,CAAC;oBACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;wBACZ,EAAE,CAAC,OAAO,CAAC,8EAA8E,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;oBACnH,CAAC;gBACH,CAAC;gBAAC,MAAM,CAAC;oBACP,EAAE,CAAC,OAAO,CAAC,8EAA8E,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;gBACnH,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC,CAAC,sBAAsB,CAAC,CAAC;IACpC,CAAC,EAAE,MAAM,CAAC,CACX,CAAC;IAEF,mEAAmE;IACnE,WAAW,CAAC,EAAE,MAAM,EAAE,oBAAoB,EAAE,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QAC5E,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAG,GAAa,CAAC,OAAO,CAAC,CAAC;IAC3E,CAAC,CAAC,CAAC;IAEH,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;AAC/C,CAAC;AAED,qFAAqF;AACrF,MAAM,CAAC,KAAK,UAAU,uBAAuB;IAC3C,KAAK,MAAM,EAAE,IAAI,gBAAgB;QAAE,aAAa,CAAC,EAAE,CAAC,CAAC;IACrD,gBAAgB,CAAC,MAAM,GAAG,CAAC,CAAC;IAC5B,MAAM,eAAe,EAAE,CAAC;AAC1B,CAAC"}

package/dist/coordination/routes.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"routes.d.ts","sourceRoot":"","sources":["../../src/coordination/routes.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAsCxD,wBAAgB,0BAA0B,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,EAAE,OAAO,aAAa,EAAE,oBAAoB,GAAG,IAAI,CA40DxK"}
+{"version":3,"file":"routes.d.ts","sourceRoot":"","sources":["../../src/coordination/routes.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AA6ExD,wBAAgB,0BAA0B,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,EAAE,OAAO,aAAa,EAAE,oBAAoB,GAAG,IAAI,CAu9DxK"}

package/dist/coordination/routes.js

@@ -16,6 +16,18 @@ function ts() {
 function coordLog(msg) {
     console.log(`${ts()} [coord] ${msg}`);
 }
+function createChannelMetrics() {
+    return {
+        attempts: 0,
+        delivered: 0,
+        failed_http: 0,
+        failed_unreachable: 0,
+        no_session: 0,
+        fallback_mailbox: 0,
+        session_disconnects: 0,
+        started_at: Date.now(),
+    };
+}
 /**
  * Optional session-token check.
  * If X-Session-Token header is present and doesn't match the stored token → returns false (caller should 403).
@@ -31,6 +43,8 @@ function sessionTokenOk(db, agentId, req) {
     return row.session_token === provided;
 }
 export function registerCoordinationRoutes(app, db, store, eventBus) {
+    // Channel push telemetry — process-scoped counters. See ChannelMetrics docs above.
+    const channelMetrics = createChannelMetrics();
     // Request logging — one line per request with method, url, status, response time
     app.addHook('onRequest', async (request) => {
         request._startTime = Date.now();
@@ -100,7 +114,12 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
             const wasDead = existing.status === 'dead';
             // Issue a fresh token on reconnect; reuse existing token for live heartbeats
             const sessionToken = wasDead ? randomUUID() : (db.prepare(`SELECT session_token FROM coord_agents WHERE id = ?`).get(existing.id).session_token ?? randomUUID());
-            db.prepare(`UPDATE coord_agents SET last_seen = datetime('now'), status = CASE WHEN status = 'dead' THEN 'idle' ELSE status END, pid = COALESCE(?, pid), capabilities = COALESCE(?, capabilities), workspace = COALESCE(?, workspace), session_token = ? WHERE id = ?`).run(pid ?? null, capsJson, workspace ?? null, sessionToken, existing.id);
+            // role IS updated on every checkin — agents know their own role and
+            // re-registrations may correct stale role values (e.g., when an old
+            // coord_agents row was inserted with role='orchestrator' before the
+            // 'coordinator' role was canonical, or when the channel-server's
+            // hardcoded role='worker' overwrote a real role).
+            db.prepare(`UPDATE coord_agents SET last_seen = datetime('now'), status = CASE WHEN status = 'dead' THEN 'idle' ELSE status END, role = ?, pid = COALESCE(?, pid), capabilities = COALESCE(?, capabilities), workspace = COALESCE(?, workspace), session_token = ? WHERE id = ?`).run(role, pid ?? null, capsJson, workspace ?? null, sessionToken, existing.id);
             const eventType = wasDead ? 'reconnected' : 'heartbeat';
             const detail = wasDead ? `${name} reconnected (was dead)` : `heartbeat from ${name}`;
             db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, ?, ?)`).run(existing.id, eventType, detail);
@@ -392,6 +411,10 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
                     db.prepare(`UPDATE coord_channel_sessions SET last_push_at = datetime('now'), push_count = push_count + 1 WHERE agent_id = ?`).run(agentId);
                 }
             }
+            else {
+                // Session disappeared between intent record and delivery — race or rapid disconnect
+                channelMetrics.no_session++;
+            }
         }
         // Bridge context to AWM engrams (outside transaction — engram store has its own DB)
         if (store && context) {
@@ -979,17 +1002,28 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
     app.get('/workers', async (req, reply) => {
         const q = workersQuerySchema.safeParse(req.query);
         const { capability, status: filterStatus, workspace } = q.success ? q.data : { capability: undefined, status: undefined, workspace: undefined };
+        // Join with coord_channel_sessions so the coordinator agent can compute
+        // alive=true for workers that have a connected channel session even when
+        // their /pulse is stale. Without this, /workers under-reports liveness
+        // during long tool-call sequences where the worker is processing but
+        // hasn't called /pulse for >5min — leading to false-positive duplicate
+        // spawns. Channel sessions get probed every 60s (coordination/index.ts:111),
+        // so a stale channel-server.js gets status='disconnected' within 60-120s.
         let workers = workspace
-            ? db.prepare(`SELECT id, name, role, status, current_task, capabilities, workspace, last_seen,
-                  ROUND((julianday('now') - julianday(last_seen)) * 86400) AS seconds_since_seen
-           FROM coord_agents
-           WHERE status != 'dead' AND role NOT IN ('orchestrator', 'coordinator') AND workspace = ?
-           ORDER BY name LIMIT 200`).all(workspace)
-            : db.prepare(`SELECT id, name, role, status, current_task, capabilities, workspace, last_seen,
-                  ROUND((julianday('now') - julianday(last_seen)) * 86400) AS seconds_since_seen
-           FROM coord_agents
-           WHERE status != 'dead' AND role NOT IN ('orchestrator', 'coordinator')
-           ORDER BY name LIMIT 200`).all();
+            ? db.prepare(`SELECT a.id, a.name, a.role, a.status, a.current_task, a.capabilities, a.workspace, a.last_seen,
+                  ROUND((julianday('now') - julianday(a.last_seen)) * 86400) AS seconds_since_seen,
+                  cs.status AS channel_status, cs.last_push_at AS channel_last_push
+           FROM coord_agents a
+           LEFT JOIN coord_channel_sessions cs ON cs.agent_id = a.id
+           WHERE a.status != 'dead' AND a.role NOT IN ('orchestrator', 'coordinator') AND a.workspace = ?
+           ORDER BY a.name LIMIT 200`).all(workspace)
+            : db.prepare(`SELECT a.id, a.name, a.role, a.status, a.current_task, a.capabilities, a.workspace, a.last_seen,
+                  ROUND((julianday('now') - julianday(a.last_seen)) * 86400) AS seconds_since_seen,
+                  cs.status AS channel_status, cs.last_push_at AS channel_last_push
+           FROM coord_agents a
+           LEFT JOIN coord_channel_sessions cs ON cs.agent_id = a.id
+           WHERE a.status != 'dead' AND a.role NOT IN ('orchestrator', 'coordinator')
+           ORDER BY a.name LIMIT 200`).all();
         if (capability) {
             workers = workers.filter(w => {
                 if (!w.capabilities)
@@ -1016,7 +1050,14 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
             workspace: w.workspace,
             lastSeen: w.last_seen,
             secondsSinceSeen: w.seconds_since_seen,
-            alive: w.seconds_since_seen < 300,
+            // alive = recent /pulse OR connected channel session.
+            // Channel sessions get probed every 60s and marked 'disconnected'
+            // when unreachable, so a connected session is reliable proof of life
+            // even during long tool sequences where the worker hasn't pulsed.
+            // Prevents duplicate worker spawns when /pulse is stale but worker is busy.
+            alive: w.seconds_since_seen < 300 || w.channel_status === 'connected',
+            channelStatus: w.channel_status,
+            channelLastPush: w.channel_last_push,
         }));
         return reply.send({
             count: result.length,
@@ -1220,6 +1261,26 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
         lines.push('# HELP coord_uptime_seconds Seconds since first agent registered');
         lines.push('# TYPE coord_uptime_seconds gauge');
         lines.push(`coord_uptime_seconds ${uptime}`);
+        // ─── Channel push telemetry (process-scoped, reset on restart) ───
+        lines.push('# HELP coord_channel_push_attempts_total Total channel push attempts since coordinator startup');
+        lines.push('# TYPE coord_channel_push_attempts_total counter');
+        lines.push(`coord_channel_push_attempts_total ${channelMetrics.attempts}`);
+        lines.push('# HELP coord_channel_push_delivered_total Successful channel deliveries');
+        lines.push('# TYPE coord_channel_push_delivered_total counter');
+        lines.push(`coord_channel_push_delivered_total ${channelMetrics.delivered}`);
+        lines.push('# HELP coord_channel_push_failed_total Failed channel deliveries by reason');
+        lines.push('# TYPE coord_channel_push_failed_total counter');
+        lines.push(`coord_channel_push_failed_total{reason="http"} ${channelMetrics.failed_http}`);
+        lines.push(`coord_channel_push_failed_total{reason="unreachable"} ${channelMetrics.failed_unreachable}`);
+        lines.push('# HELP coord_channel_no_session_total Push attempts where agent had no connected session');
+        lines.push('# TYPE coord_channel_no_session_total counter');
+        lines.push(`coord_channel_no_session_total ${channelMetrics.no_session}`);
+        lines.push('# HELP coord_channel_fallback_mailbox_total Pushes that fell back to mailbox after delivery failure');
+        lines.push('# TYPE coord_channel_fallback_mailbox_total counter');
+        lines.push(`coord_channel_fallback_mailbox_total ${channelMetrics.fallback_mailbox}`);
+        lines.push('# HELP coord_channel_session_disconnects_total Sessions marked disconnected after delivery failure');
+        lines.push('# TYPE coord_channel_session_disconnects_total counter');
+        lines.push(`coord_channel_session_disconnects_total ${channelMetrics.session_disconnects}`);
         return reply.type('text/plain; version=0.0.4; charset=utf-8').send(lines.join('\n') + '\n');
     });
     // ─── Deep Health ───────────────────────────────────────────────
@@ -1302,6 +1363,7 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
      * Returns { delivered, error? }. On connection failure, marks session dead.
      */
     async function deliverToChannel(agentId, channelUrl, content, meta) {
+        channelMetrics.attempts++;
         try {
             const res = await fetch(`${channelUrl}/push`, {
                 method: 'POST',
@@ -1310,24 +1372,54 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
                 signal: AbortSignal.timeout(5000),
             });
             if (!res.ok) {
+                channelMetrics.failed_http++;
                 return { delivered: false, error: `channel returned ${res.status}` };
             }
+            channelMetrics.delivered++;
             return { delivered: true };
         }
         catch (err) {
             // Connection refused / timeout → worker process is dead, mark session disconnected
+            channelMetrics.failed_unreachable++;
+            channelMetrics.session_disconnects++;
             db.prepare(`UPDATE coord_channel_sessions SET status = 'disconnected' WHERE agent_id = ?`).run(agentId);
             const agent = db.prepare(`SELECT name FROM coord_agents WHERE id = ?`).get(agentId);
             coordLog(`channel/deliver FAILED → ${agent?.name ?? agentId}: ${err instanceof Error ? err.message : err} — session marked disconnected`);
             return { delivered: false, error: `worker unreachable: ${err instanceof Error ? err.message : err}` };
         }
     }
-    /** POST /channel/push — Push a message to an agent. Tries live delivery first, falls back to mailbox queue. */
+    /** POST /channel/push — Push a message to an agent. Tries live delivery first, falls back to mailbox queue.
+     *
+     * Two addressing modes:
+     *   - {agentId, message}                — direct UUID
+     *   - {role, workspace, message}        — server resolves to most-recently-seen alive agent
+     *                                          matching role+workspace. Used by workers to notify
+     *                                          coordinator (whose UUID changes across restarts).
+     */
     app.post('/channel/push', async (request, reply) => {
         const parsed = channelPushSchema.safeParse(request.body);
         if (!parsed.success)
             return reply.status(400).send({ error: parsed.error.flatten() });
-        const { agentId, message } = parsed.data;
+        const { message } = parsed.data;
+        let { agentId } = parsed.data;
+        // Role-based addressing — resolve to a concrete agentId
+        if (!agentId && parsed.data.role && parsed.data.workspace) {
+            const resolved = db.prepare(`SELECT id FROM coord_agents
+         WHERE role = ? AND workspace = ? AND status != 'dead'
+         ORDER BY last_seen DESC
+         LIMIT 1`).get(parsed.data.role, parsed.data.workspace);
+            if (!resolved) {
+                return reply.status(404).send({
+                    error: `No alive agent found for role='${parsed.data.role}' workspace='${parsed.data.workspace}'`,
+                });
+            }
+            agentId = resolved.id;
+        }
+        // Type narrowing — Zod refine guarantees agentId is set by this point,
+        // but TypeScript can't see through the refine. This guard is unreachable
+        // in practice (would have 400'd earlier).
+        if (!agentId)
+            return reply.status(400).send({ error: 'Internal: agentId resolution failed' });
         const agent = db.prepare(`SELECT name, workspace FROM coord_agents WHERE id = ?`).get(agentId);
         if (!agent)
             return reply.status(404).send({ error: 'Agent not found' });
@@ -1342,6 +1434,11 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
                 return reply.send({ ok: true, delivered: true, channelId: session.channel_id });
             }
             // Live delivery failed — fall through to mailbox
+            channelMetrics.fallback_mailbox++;
+        }
+        else {
+            // No connected session — push went straight to mailbox
+            channelMetrics.no_session++;
         }
         // Queue to mailbox (delivered on next /next poll)
         db.prepare(`INSERT INTO coord_mailbox (worker_name, workspace, message, source) VALUES (?, ?, ?, 'coordinator')`).run(agent.name, agent.workspace, message);
@@ -1392,5 +1489,46 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
             coordLog(`channel/probe: ${alive} alive, ${dead} dead — dead sessions marked disconnected`);
         return reply.send({ probed: results.length, alive, dead, results });
     });
+    /**
+     * GET /telemetry/channels — Channel push delivery telemetry.
+     *
+     * Counters reset on coordinator restart (in-process). Use this to answer:
+     *   "Are channels reliable enough to depend on, or do we need a polling fallback?"
+     *
+     * Response shape:
+     *   {
+     *     since: ISO timestamp of when counters started,
+     *     uptime_seconds: number,
+     *     attempts, delivered, failed_http, failed_unreachable,
+     *     no_session, fallback_mailbox, session_disconnects: number,
+     *     delivery_rate: 0..1 (delivered / attempts) or null if zero attempts,
+     *     per_agent: [{ agent_name, push_count, last_push_at, status }]
+     *   }
+     */
+    app.get('/telemetry/channels', async (_request, reply) => {
+        const perAgent = db.prepare(`
+      SELECT a.name AS agent_name, cs.push_count, cs.last_push_at, cs.status,
+             cs.connected_at
+      FROM coord_channel_sessions cs
+      JOIN coord_agents a ON a.id = cs.agent_id
+      ORDER BY cs.push_count DESC, cs.connected_at DESC
+    `).all();
+        const deliveryRate = channelMetrics.attempts > 0
+            ? channelMetrics.delivered / channelMetrics.attempts
+            : null;
+        return reply.send({
+            since: new Date(channelMetrics.started_at).toISOString(),
+            uptime_seconds: Math.round((Date.now() - channelMetrics.started_at) / 1000),
+            attempts: channelMetrics.attempts,
+            delivered: channelMetrics.delivered,
+            failed_http: channelMetrics.failed_http,
+            failed_unreachable: channelMetrics.failed_unreachable,
+            no_session: channelMetrics.no_session,
+            fallback_mailbox: channelMetrics.fallback_mailbox,
+            session_disconnects: channelMetrics.session_disconnects,
+            delivery_rate: deliveryRate,
+            per_agent: perAgent,
+        });
+    });
 }
 //# sourceMappingURL=routes.js.map