Skip to content

claude-flow

v3.10.31 Bugfix

This release fixes issues for SREs watching stability and regressions.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agentic-ai agentic-framework agentic-rag agentic-workflow agents ai-agents
+14 more
ai-assistant ai-coding ai-skills autonomous-agents claude-code codex mcp-server multi-agent multi-agent-systems npm skills swarm swarm-intelligence typescript

Summary

AI summary

Updates Bug fixes, @grym3s, and @shaharKeisarApps across a mixed release.

Changes in this release

Feature Low

Adds env variable `CLAUDE_FLOW_MAX_UNCERTAINTY` to override maximum uncertainty threshold.

Adds env variable `CLAUDE_FLOW_MAX_UNCERTAINTY` to override maximum uncertainty threshold.

Source: llm_adapter@2026-05-31

Confidence: high

Feature Low

Adds four regression tests covering router behavior, cold-start preservation, env override, and daemon promise handling.

Adds four regression tests covering router behavior, cold-start preservation, env override, and daemon promise handling.

Source: llm_adapter@2026-05-31

Confidence: high

Bugfix Medium

Model router escalation now skips when target mean is worse than selected or selected has confident posterior.

Model router escalation now skips when target mean is worse than selected or selected has confident posterior.

Source: llm_adapter@2026-05-31

Confidence: high

Bugfix Medium

Daemon worker trigger now awaits headless executor initialization promise to prevent race condition.

Daemon worker trigger now awaits headless executor initialization promise to prevent race condition.

Source: llm_adapter@2026-05-31

Confidence: high

Full changelog

Bug fixes

#2250 — model router escalation overrode Thompson bandit

Trained opus-suppressed priors were discarded by an "escalate one tier up" step that fired on ~every trivial route (the uncertainty formula was structurally >0.6, gate at 0.15). Reporter (@grym3s) measured opus selected ~40% on trivial tasks, haiku never selected.

Fix in v3/@claude-flow/cli/src/ruvector/model-router.ts:

  • Escalation now skips when (a) target mean is meaningfully worse than selected (< selected_mean − 0.10), OR (b) selected has confident+decent posterior (α+β ≥ 5 AND mean ≥ 0.45)
  • Cold-start Beta(1,1) priors fail both checks → unlearned routers preserve original conservative behavior
  • New env override: CLAUDE_FLOW_MAX_UNCERTAINTY=0.0..1.0

#2251 — daemon trigger -w <worker> raced headless init

Constructor kicked off initHeadlessExecutor() fire-and-forget. On-demand daemon trigger called triggerWorker() immediately after construction, raced the init, and fell through to the local stub in ~2ms. Scheduled fires were unaffected (long-running daemon never raced). Reporter (@shaharKeisarApps).

Fix in v3/@claude-flow/cli/src/services/worker-daemon.ts:

  • New headlessInitPromise field capturing the constructor's init promise
  • triggerWorker() awaits it before checking headlessAvailable

Regression guard

v3/@claude-flow/cli/__tests__/issue-2250-2251-regression.test.ts — 4 new tests:

  • Loads reporter's exact suppressed-opus prior, asserts opus rate <20% AND haiku selected ≥1 time over 50 trivial routes
  • Cold-start preservation
  • Env override
  • Deferred-promise harness proves executeWorker() cannot fire until headlessInitPromise resolves

All 12 router + daemon tests pass. TypeScript clean.

Install

npx ruflo@latest                # or @alpha / @v3alpha
npx @claude-flow/cli@latest
npx claude-flow@latest

All three packages published to npm, all three dist-tags (latest, alpha, v3alpha) pointing at 3.10.31.

What's NOT in this release

5-dataset BEIR benchmark work (TREC-COVID ingest still in flight) → ships as 3.10.32.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track claude-flow

Get notified when new releases ship.

Sign up free

About claude-flow

Deploy multi-agent swarms with coordinated workflows.

All releases →

Related context

Beta — feedback welcome: [email protected]