claude-flow 3.38.15 → 3.38.18

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-flow",
-  "version": "3.38.15",
+  "version": "3.38.18",
   "workspaces": [
     "v3/@claude-flow/codex",
     "v3/@claude-flow/plugin-agent-federation",

package/v3/@claude-flow/cli/catalog-manifest.json

@@ -1,8 +1,8 @@
 {
   "schemaVersion": 1,
   "generation": 5,
-  "generatedAt": "2026-08-21T14:40:50.895Z",
-  "gitSha": "87bacd55",
+  "generatedAt": "2026-08-22T16:23:54.994Z",
+  "gitSha": "c3132764",
   "catalog": {
     "agents": 165,
     "tools": 397,

package/v3/@claude-flow/cli/dist/src/commands/doctor.js

@@ -506,6 +506,134 @@ async function checkMemoryStructuralIntegrity() {
         catch { /* best-effort */ }
     }
 }
+// #2968 option 1 — read-only doctor check for the active SQLite driver.
+//
+// Native better-sqlite3 (durable, WAL-capable) can silently degrade to the
+// sql.js WASM fallback (non-durable — `wal_checkpoint` calls are rejected
+// and the write is lost) when a postinstall script is skipped. `memory
+// store` still printed "Data stored successfully" before the persistWarning
+// fix in #2983/3.38.1 (see memory-store-persist-warning-2968.test.ts); this
+// check gives a standing, read-only signal so the driver split is visible
+// any time doctor runs, independent of any single store call.
+//
+// Table count in the on-disk memory.db is the cheap, reliable signal from
+// the issue report: the native driver's schema produces 47 tables, the
+// sql.js fallback's produces only 10. Deliberately does NOT change install
+// behavior (that's option 2 from #2968, explicitly out of scope here) —
+// this only reports, it never repairs.
+const MEMORY_DRIVER_NATIVE_TABLE_FLOOR = 20; // roughly midpoint of sql.js's ~10 and native's ~47
+async function checkMemoryPersistenceDriver() {
+    const NAME = 'Memory Persistence Driver';
+    const dbPath = await resolveMemoryDbPath();
+    if (!dbPath) {
+        return {
+            name: NAME,
+            status: 'warn',
+            message: 'no memory.db found (see Memory Database Presence above) — driver check skipped',
+        };
+    }
+    if (isMemoryDbEncryptedAtRest(dbPath)) {
+        return {
+            name: NAME,
+            status: 'warn',
+            message: `${dbPath} — RFE1-encrypted at rest; driver check can't run without decrypting (expected, not corruption)`,
+        };
+    }
+    let Database;
+    try {
+        Database = (await import('better-sqlite3')).default;
+    }
+    catch {
+        Database = null;
+    }
+    let tableCount = null;
+    let nativeUnavailableReason = null;
+    let nativeOpenOtherError = null;
+    if (Database) {
+        let db;
+        try {
+            db = new Database(dbPath, { readonly: true, fileMustExist: true });
+        }
+        catch (e) {
+            if (isNativeSqliteBindingUnavailable(e)) {
+                nativeUnavailableReason = (e.message || String(e)).split(/\r?\n/, 1)[0];
+            }
+            else {
+                nativeOpenOtherError = e.message || String(e);
+            }
+            db = null;
+        }
+        if (db) {
+            try {
+                const row = db.prepare("SELECT count(*) AS c FROM sqlite_master WHERE type='table'").get();
+                tableCount = Number(row?.c ?? 0);
+            }
+            catch {
+                // leave null — Memory Integrity above already reports open/query failures
+            }
+            finally {
+                try {
+                    db.close();
+                }
+                catch { /* best-effort */ }
+            }
+        }
+    }
+    else {
+        nativeUnavailableReason = 'better-sqlite3 not installed';
+    }
+    // Fall back to sql.js purely to report table count when the native path
+    // couldn't — informational only, never used to claim durability.
+    if (tableCount === null && (nativeUnavailableReason || nativeOpenOtherError)) {
+        const sdb = await tryOpenSqlJs(dbPath);
+        if (sdb) {
+            try {
+                const res = sdb.exec("SELECT count(*) FROM sqlite_master WHERE type='table'");
+                tableCount = Number(res[0]?.values?.[0]?.[0] ?? 0);
+            }
+            catch {
+                // leave null
+            }
+            finally {
+                try {
+                    sdb.close();
+                }
+                catch { /* best-effort */ }
+            }
+        }
+    }
+    const tableSummary = tableCount === null
+        ? 'table count unavailable'
+        : `${tableCount} tables (native schema ~47, sql.js-fallback schema ~10 — #2968)`;
+    if (nativeUnavailableReason) {
+        return {
+            name: NAME,
+            status: 'warn',
+            message: `${dbPath} — active driver: sql.js (WASM fallback, non-durable) — native better-sqlite3 binding unavailable: ${nativeUnavailableReason} — wal_checkpoint calls silently no-op, writes may not persist across processes (#2968/#2867/#2219) [${tableSummary}]`,
+            fix: 'reinstall with npm install scripts enabled, or run `npm rebuild better-sqlite3`; then rerun this check',
+        };
+    }
+    if (nativeOpenOtherError) {
+        return {
+            name: NAME,
+            status: 'warn',
+            message: `${dbPath} — native better-sqlite3 module loadable, but could not open this database (${nativeOpenOtherError}) — see Memory Integrity above for the corruption/encryption diagnosis [${tableSummary}]`,
+        };
+    }
+    if (tableCount !== null && tableCount < MEMORY_DRIVER_NATIVE_TABLE_FLOOR) {
+        return {
+            name: NAME,
+            status: 'warn',
+            message: `${dbPath} — active driver: native better-sqlite3, but this database has only ${tableCount} tables — that matches the sql.js-fallback schema shape (~10), not the native schema (~47); it was likely created before the native binding became available, and durable writes made before then may be missing`,
+            fix: 'back up .swarm/memory.db then `claude-flow memory init --force` to rebuild under the native driver',
+        };
+    }
+    return {
+        name: NAME,
+        status: 'pass',
+        message: `${dbPath} — active driver: native better-sqlite3 (durable, WAL-capable) [${tableSummary}]`,
+    };
+}
 // Check 1 (--component memory, deep path) — #2737 part 2 strengthens this:
 // prefer native better-sqlite3 PRAGMA integrity_check (adds index↔table
 // cross-checking + UNIQUE verification that quick_check above deliberately
@@ -2220,6 +2348,7 @@ export const doctorCommand = {
             checkDaemonStatus,
             checkMemoryDatabase,
             checkMemoryStructuralIntegrity, // #2737 — bounded, native quick_check on every default run
+            checkMemoryPersistenceDriver, // #2968 — native better-sqlite3 vs sql.js fallback, read-only
             checkLearningBridge, // #2545 — can the auto-memory hook actually load @claude-flow/memory?
             checkApiKeys,
             checkMcpServers,
@@ -2257,6 +2386,7 @@ export const doctorCommand = {
             'memory': [
                 checkMemoryDatabase, // existing: exists + statable (unchanged)
                 checkMemoryIntegrity, // #2677 check 1: sql.js open + PRAGMA integrity_check
+                checkMemoryPersistenceDriver, // #2968: native better-sqlite3 vs sql.js fallback
                 checkMemoryContent, // #2677 check 2: memory_entries content coverage
                 checkMemoryEmbeddingCoverage, // #2677 check 3: vector coverage on populated rows
                 checkMemoryReflexionCoverage, // #2677 check 6: episodes are retrievable

package/v3/@claude-flow/cli/dist/src/ruvector/model-router.d.ts

@@ -77,6 +77,18 @@ export interface ModelRouterConfig {
     enableCostOptimization: boolean;
     /** Prefer faster models when confidence is high (default: true) */
     preferSpeed: boolean;
+    /**
+     * Discounted Thompson sampling factor applied to every bucket/model's
+     * α/β on each recordOutcome call, BEFORE that call's own reward is added
+     * (Discounted TS — arXiv 2305.10718): `α ← γα`, `β ← γβ`. Range (0, 1];
+     * 1 = disabled (default — fully backward compatible with persisted
+     * `.swarm/model-router-state.json` state). γ<1 makes old routing history
+     * decay geometrically so the bandit recovers faster after a sustained
+     * shift in a model's real-world quality, instead of unbounded accumulated
+     * history permanently dominating the posterior. Mirrors the epsilon decay
+     * `q-learning-router.ts` already applies to its own exploration rate.
+     */
+    priorDecay: number;
 }
 /**
  * Routing decision result

package/v3/@claude-flow/cli/dist/src/ruvector/model-router.js

@@ -305,7 +305,13 @@ const DEFAULT_CONFIG = {
     autoSaveInterval: 1, // Save after every decision for CLI persistence
     enableCostOptimization: true,
     preferSpeed: true,
+    priorDecay: 1,
 };
+// Beta shape parameters must stay well clear of 0 for sampleGamma/sampleBeta
+// to remain numerically stable across a persisted, long-running state file —
+// see priorDecay's doc comment. Geometric decay alone (no reward added back)
+// would otherwise shrink an unused bucket/model's α,β toward 0 forever.
+const PRIOR_DECAY_FLOOR = 0.05;
 // Posterior mean of a Beta(α,β) prior — used by the #2250 escalation guard
 // to detect when the bandit has *learned* the escalation target is worse.
 function priorMean(p) {
@@ -346,6 +352,17 @@ export class ModelRouter {
     abComparisons = 0;
     constructor(config = {}) {
         this.config = { ...DEFAULT_CONFIG, ...config };
+        // priorDecay multiplies directly into persisted α/β (recordOutcome). An
+        // out-of-range value (NaN, <=0, >1 — e.g. a bad env/config parse) would
+        // silently poison `.swarm/model-router-state.json` forever: Math.max's
+        // NaN-poisoning bypasses sampleBeta's own alpha<=0||beta<=0 fallback, and
+        // a negative value pins every prior at PRIOR_DECAY_FLOOR on the very
+        // first call. Fall through to the safe default (1 = disabled) instead of
+        // throwing, matching this file's existing malformed-config idiom (see
+        // envMaxUncertainty above).
+        if (!Number.isFinite(this.config.priorDecay) || this.config.priorDecay <= 0 || this.config.priorDecay > 1) {
+            this.config.priorDecay = 1;
+        }
         this.state = this.loadState();
     }
     /**
@@ -897,6 +914,21 @@ export class ModelRouter {
         // is wasteful even when correct). Failure/escalation always β++.
         if (!this.state.priors)
             this.state.priors = defaultBucketedPriors();
+        // Discounted Thompson sampling (priorDecay < 1): decay ALL bucket/model
+        // priors once per recordOutcome call, before this call's own reward is
+        // added — every routing decision is one discrete "time step" for every
+        // arm, not just the one that got pulled, matching the arXiv 2305.10718
+        // formulation. No-op when priorDecay is the default of 1.
+        if (this.config.priorDecay < 1) {
+            for (const bk of Object.keys(this.state.priors)) {
+                const bucketPriors = this.state.priors[bk];
+                for (const m of Object.keys(bucketPriors)) {
+                    const p = bucketPriors[m];
+                    p.alpha = Math.max(PRIOR_DECAY_FLOOR, p.alpha * this.config.priorDecay);
+                    p.beta = Math.max(PRIOR_DECAY_FLOOR, p.beta * this.config.priorDecay);
+                }
+            }
+        }
         const bp = this.state.priors[bucket] ?? (this.state.priors[bucket] = defaultBanditPriors());
         const reward = BANDIT_REWARDS[model]?.[outcome] ?? 0.5;
         bp[model].alpha += reward;

package/v3/@claude-flow/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@claude-flow/cli",
-  "version": "3.38.15",
+  "version": "3.38.18",
   "type": "module",
   "description": "Ruflo CLI - Enterprise AI agent orchestration with 60+ specialized agents, swarm coordination, MCP server, self-learning hooks, and vector memory for Claude Code",
   "main": "dist/src/index.js",
@@ -93,6 +93,7 @@
     "publish:all": "./scripts/publish.sh"
   },
   "devDependencies": {
+    "tsx": "^4.19.0",
     "typescript": "^5.3.0",
     "vitest": "^4.1.0"
   },
@@ -112,10 +113,10 @@
     "chalk": "^5.3.0",
     "commander": "^12.0.0",
     "cors": "^2.8.5",
-    "fs-extra": "^11.2.0",
     "express": ">=4.22.2",
     "express-rate-limit": ">=8.4.1",
     "fast-uri": "3.1.5",
+    "fs-extra": "^11.2.0",
     "helmet": "^7.2.0",
     "inquirer": "^9.2.0",
     "semver": "7.7.3",
@@ -127,12 +128,12 @@
   },
   "optionalDependencies": {
     "@agntcy/slim-bindings": "2.0.0-alpha.5",
-    "@napi-rs/keyring": "1.3.0",
-    "@claude-flow/memory": "^3.0.0-alpha.22",
+    "@claude-flow/memory": "^3.0.0-alpha.23",
     "@metaharness/darwin": "~0.9.0",
     "@metaharness/flywheel": "~0.1.10",
     "@metaharness/radio": "~0.1.0",
     "@metaharness/turn-credit": "~0.1.0",
+    "@napi-rs/keyring": "1.3.0",
     "agentdb": "^3.0.0-alpha.17",
     "agentic-flow": "^3.0.0-alpha.1",
     "better-sqlite3": "^12.9.0",

package/.claude/coordinator/config.json

@@ -1,24 +0,0 @@
-{
-  "version": "1.0",
-  "startedAt": "2026-07-08T02:30:00Z",
-  "policy": "full-autopilot",
-  "safety": {
-    "requireCiGreen": true,
-    "requireNoConflict": true,
-    "requireIdleHours": 0,
-    "maxTicks": 30,
-    "tickIntervalSeconds": 600,
-    "excludeIfCodeownersBlock": true
-  },
-  "targetPRs": [
-    {"num": 2517, "adr": null, "cluster": "security", "grade": "A", "note": "AgentDB retrieval poisoning 93-100% ASR"},
-    {"num": 2537, "adr": null, "cluster": "memory", "grade": "B", "note": "AutoMem RL loop — needs ADR renumber (collides w/ #2527 on 167)"},
-    {"num": 2577, "adr": 176, "cluster": "performance", "grade": "A", "note": "PolyKV shared KV pool 97.7% reduction"},
-    {"num": 2589, "adr": null, "cluster": "security", "grade": "A", "note": "VMG + RepE IPI — needs ADR renumber (collides w/ #2598 on 178)"},
-    {"num": 2598, "adr": null, "cluster": "intelligence", "grade": "B", "note": "SkillRL — needs ADR renumber (collides w/ #2589 on 178)"}
-  ],
-  "adrCollisions": [
-    {"conflictingPRs": [2589, 2598], "originalAdr": 178, "resolution": "earlier-wins", "reason": "2589 (dream 2026-07-06) beats 2598 (dream 2026-07-07)"},
-    {"conflictingPRs": [2527, 2537], "originalAdr": 167, "resolution": "earlier-wins", "reason": "2527 (dream 2026-07-02) beats 2537 (dream 2026-07-03) — but 2527 not in scope so 2537 gets it if 2527 is also renumbered"}
-  ]
-}

package/.claude/coordinator/state.json

@@ -1,37 +0,0 @@
-{
-  "tick": 5,
-  "terminated": true,
-  "terminationReason": "main-preexisting-red-cannot-autopilot",
-  "closed": [],
-  "merged": [2589],
-  "blocked": [2517, 2537, 2577, 2598],
-  "adrAssignments": {
-    "2517": 180,
-    "2537": 181,
-    "2577": 182,
-    "2598": 183,
-    "2589": 178
-  },
-  "routingWeights": {
-    "parallel-renumber-adr": -1,
-    "centralized-adr-pre-assign": 2,
-    "convert-ready-when-green": 1,
-    "single-action-per-tick": 1,
-    "gh-pr-merge-admin-squash": 1,
-    "rebase-onto-main-when-shared-blocker": 2,
-    "force-remove-stale-worktree": 1,
-    "inline-fix-when-workflow-stalls": 1,
-    "diagnose-main-before-per-pr-fix": 3,
-    "terminate-loop-when-main-red-and-not-caused-by-scope": 2
-  },
-  "cumulativeReward": 14,
-  "ticksSpent": 5,
-  "completedAt": "2026-07-08T03:15:00Z",
-  "achievements": [
-    "Landed PR #2589 (security VMG+RepE IPI) — first dream merge in weeks",
-    "Resolved 4 ADR-number collisions (178/180/181/182/183) that had built up across in-flight dreams",
-    "Rebased 4 stale dream branches onto main to inherit v3.25.3 fixes",
-    "Learned + persisted routing weights via 5 tick cycles"
-  ],
-  "handoff": "Main has 7 preexisting failing checks (Build V3 all platforms, Witness verify all platforms, graph schema smoke ADR-130 P1). All 4 remaining PRs (#2517, #2537, #2577, #2598) are draft=false + mergeable + at latest ADR numbers + rebased to main tip. They will auto-merge via a re-fired coordinator OR via `gh pr merge --admin --squash` once main goes green. Recommendation: file a `fix/main-v3-ci-red` triage issue first, unblock main, then re-fire the coordinator with `/loop 10m` OR merge the 4 PRs manually with --admin since they only add ADR markdown."
-}

package/.claude/coordinator/tick-log/tick-01.json

@@ -1,17 +0,0 @@
-{
-  "tick": 1,
-  "completedAt": "2026-07-08T02:33:00Z",
-  "durationSeconds": 141,
-  "assessed": 5,
-  "advanced": 5,
-  "merged": 0,
-  "renumbered": 4,
-  "readyForCi": 1,
-  "totalReward": 5,
-  "coordinatorMistake": {
-    "type": "parallel-race-adr-allocation",
-    "description": "4 independent assessors each picked ADR-180 as next-free. Result: 4 PRs now all claim ADR-180.",
-    "affectedPRs": [2517, 2537, 2577, 2598]
-  },
-  "lessonLearned": "Future ticks: PRE-ASSIGN ADR numbers centrally before fanning out. Never let parallel agents pick numbers independently."
-}

package/.claude/coordinator/tick-log/tick-02.json

@@ -1,17 +0,0 @@
-{
-  "tick": 2,
-  "completedAt": "2026-07-08T02:36:00Z",
-  "durationSeconds": 106,
-  "collisionFixesSucceeded": 3,
-  "advancesSucceeded": 2,
-  "mergedThisTick": [2589],
-  "renumberedThisTick": [
-    {"pr": 2537, "adr": 181},
-    {"pr": 2577, "adr": 182},
-    {"pr": 2598, "adr": 183}
-  ],
-  "ciRunningThisTick": [2517],
-  "totalReward": 5,
-  "cumulativeReward": 10,
-  "lesson": "Central pre-assignment resolved the parallel-race collision that tick 1 created. Reward pattern confirms centralized-adr-pre-assign > parallel-renumber-adr."
-}

package/.claude/coordinator/tick-log/tick-03.json

@@ -1,17 +0,0 @@
-{
-  "tick": 3,
-  "completedAt": "2026-07-08T02:59:00Z",
-  "durationSeconds": 69,
-  "assessed": 4,
-  "advanced": 0,
-  "merged": 0,
-  "totalReward": 0,
-  "cumulativeReward": 10,
-  "sharedBlocker": {
-    "type": "old-base-missing-2578-fix",
-    "description": "All 4 remaining dream branches predate the #2578 phantom-subpath fix that landed in v3.25.3 (PR #2602). Their old midstream-aware-loader.ts still imports from 'agentic-flow/transport/loader' → TS errors on Build V3.",
-    "affectedPRs": [2517, 2537, 2577, 2598],
-    "resolution": "rebase all 4 onto latest main; the fix is already there"
-  },
-  "lessonLearned": "Old-base branches get free fixes on rebase after main advances. Detect the shared-blocker pattern (same failure mode across N PRs) → try rebase before per-PR fixes."
-}

package/.claude/coordinator/tick-log/tick-04.json

@@ -1,14 +0,0 @@
-{
-  "tick": 4,
-  "completedAt": "2026-07-08T03:03:00Z",
-  "durationSeconds": 92,
-  "rebased": [2517, 2537, 2577, 2598],
-  "convertedReady": [2537, 2577, 2598],
-  "coordinatorMistakes": [
-    {"type": "schema-drift", "description": "3 rebase agents returned prNum=0 instead of the target PR number; downstream Convert-Ready phase failed on nonexistent PR #0"},
-    {"type": "stale-worktree", "description": "Tick 1's worktree wf_ea2955a6-e74-10 held dream/2026-07-07-intelligence; tick 4's #2598 agent couldn't checkout. Fixed inline with git worktree remove --force + inline rebase."}
-  ],
-  "totalReward": 4,
-  "cumulativeReward": 14,
-  "lesson": "Workflow agents must inject their prNum into the return payload (prompt-side templating). Force-remove worktrees at end of prior workflow OR clean up before spawning new ones. Sometimes inline main-context work is faster than another workflow round."
-}

package/.claude/coordinator/tick-log/tick-05.json

@@ -1,16 +0,0 @@
-{
-  "tick": 5,
-  "completedAt": "2026-07-08T03:15:00Z",
-  "durationSeconds": 33,
-  "assessed": 4,
-  "mergeEligible": 0,
-  "merged": 0,
-  "totalReward": 0,
-  "cumulativeReward": 14,
-  "terminalDiagnosis": {
-    "type": "main-preexisting-red",
-    "description": "Main's latest v3-ci run (sha a444930d8, post-#2589 merge) has 7 failing checks preexisting: Build V3 (macos/ubuntu/windows) [TS type errors in plugin-agent-federation], Witness verify (signed manifest) [macos/ubuntu/windows], graph schema smoke (ADR-130 P1). ALL 4 remaining dream PRs inherit these failures. They can NEVER pass CI while main is red.",
-    "canCoordinatorFixThis": false,
-    "reasoning": "Multi-platform TS type errors + witness manifest issues + graph schema audit = multi-hour repo-level work that needs human architectural decisions (revert vs fix upstream, which shim signatures to accept). Beyond the coordinator's 10-min autopilot scope."
-  }
-}