claude-flow

v3.10.17 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 4d AI Agents & Assistants

View tool

✓ 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 What's in the box, Honest limits, and ADR-077 across a mixed release.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	Adds pretrain-from-github script to seed trajectories, patterns, and neural store from repo history. Adds pretrain-from-github script to seed trajectories, patterns, and neural store from repo history. Source: llm_adapter@2026-05-30 Confidence: high	—
Feature	Low	Adds benchmark-pretrained-retrieval script to validate pretraining results. Adds benchmark-pretrained-retrieval script to validate pretraining results. Source: llm_adapter@2026-05-30 Confidence: high	—
Feature	Low	Adds unit test __tests__/pretrain-from-github.test.ts for CI guard. Adds unit test __tests__/pretrain-from-github.test.ts for CI guard. Source: llm_adapter@2026-05-30 Confidence: high	—
Feature	Low	Updates documentation: ADR-077-pretrain-from-history.md and learning/self-learning-usage.md. Updates documentation: ADR-077-pretrain-from-history.md and learning/self-learning-usage.md. Source: llm_adapter@2026-05-30 Confidence: high	—
Dependency	Low	Updates all packages (@claude-flow/cli, claude-flow, ruflo) to version 3.10.17 on latest, alpha, and v3alpha. Updates all packages (@claude-flow/cli, claude-flow, ruflo) to version 3.10.17 on latest, alpha, and v3alpha. Source: llm_adapter@2026-05-30 Confidence: high	—
Performance
Performance	Medium	Improves pretraining latency to ~3.36 ms per item. Improves pretraining latency to ~3.36 ms per item. Source: llm_adapter@2026-05-30 Confidence: high	—
Performance	Medium	Improves retrieval match rate to 100 % across 10 sample queries. Improves retrieval match rate to 100 % across 10 sample queries. Source: llm_adapter@2026-05-30 Confidence: high	—
Performance	Medium	Improves average query latency to 7.67 ms after pretraining. Improves average query latency to 7.67 ms after pretraining. Source: llm_adapter@2026-05-30 Confidence: high	—
Bugfix	Low	Fixes empty‑state issue where new ruflo installs reported 0 patterns and trajectories. Fixes empty‑state issue where new ruflo installs reported 0 patterns and trajectories. Source: llm_adapter@2026-05-30 Confidence: low	—

Full changelog

What ships

Pretrain self-learning from a repo's GitHub history (ADR-077). One script,
zero config, ~3 ms per trajectory. Turns the day-one "0 patterns, 0
trajectories" problem into a one-liner.

node v3/@claude-flow/cli/scripts/pretrain-from-github.mjs
# → 80 trajectories trained from 50 commits + 30 issues
# → +95 trajectoriesRecorded, +85 patternsLearned, +80 neuralPatternCount
# → 100% retrieval match rate across 10 sample queries

Why

ADR-074–076 (3.10.14–3.10.16) fixed honesty, coherence, and retrieval
quality. But a fresh ruflo install still started empty — every "did learning
happen?" call legitimately returned 0 until many real sessions had run. This
release closes that gap by seeding from a signal source every repo already has:
its own commits and issues.

Each item flows through the same code paths real-time learning uses (no
shortcuts) — distillAndSerialise (ADR-076) → recordTrajectory (ADR-074) →
neural store seed. That also closes the ADR-075 consistency note
"globalStats moved but neural_patterns stayed empty" by writing to both
stores from the same script.

Measured proof

| | Before | After | Δ |
|---|---:|---:|---:|
| trajectoriesRecorded | 0 | 95 | +95 |
| patternsLearned | 0 | 85 | +85 |
| neuralPatternCount | 15 | 95 | +80 |
| Trained / harvested | — | — | 80/80 |
| Avg pretrain latency | — | — | 3.36 ms/item |
| Retrieval match rate (N=95) | — | — | 100% (10/10) |
| Avg query latency | — | — | 7.67 ms |

Run JSONs:

docs/benchmarks/runs/pretrain-from-github-latest.json
docs/benchmarks/runs/pretrained-retrieval-latest.json

What's in the box

scripts/pretrain-from-github.mjs — env-configurable harvester
(COMMITS, ISSUES, SOURCE, BENCH_JSON)
scripts/benchmark-pretrained-retrieval.mjs — after-pretrain validator
__tests__/pretrain-from-github.test.ts — CI guard with embedded fixture
(no live git/gh in tests; auto-picked-up by v3-ci.yml)
v3/docs/adr/ADR-077-pretrain-from-history.md
v3/docs/learning/self-learning-usage.md — copy-paste guide covering all
three learning paths plus pretrain

Reproduce

git clone https://github.com/ruvnet/ruflo && cd ruflo
npm install && ( cd v3/@claude-flow/cli && npx tsc -b )
node v3/@claude-flow/cli/scripts/pretrain-from-github.mjs
node v3/@claude-flow/cli/scripts/benchmark-pretrained-retrieval.mjs
( cd v3/@claude-flow/cli && npx vitest run __tests__/pretrain-from-github.test.ts )

Honest limits

Standalone-process drift: when the script runs outside the live MCP daemon,
sonaCoordinator and memory-bridge start empty. The script's consistency
block flags this explicitly. From inside the daemon both stores are warm.
100% match-rate ≠ semantic relevance. Pretrain proves the wiring; ADR-076's
MRR benchmark is the right gauge for relevance quality.
Commits are all recorded as success (no outcome signal). A "was this commit
reverted?" verdict refinement is tracked for follow-up.

Install

npx [email protected]                 # or @latest
npx ruflo@alpha                   # legacy compat
npx ruflo@v3alpha                 # legacy compat

All three packages (@claude-flow/cli, claude-flow, ruflo) are at
3.10.17 on latest, alpha, and v3alpha.

View diff on GitHub

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

Share on X Share on Bluesky

Track claude-flow

Get notified when new releases ship.

About claude-flow

Deploy multi-agent swarms with coordinated workflows.

All releases →