v1.7.11 — X → NotebookLM ingest pipeline + TurboQuant inline

Highlights

End-to-end X → NotebookLM ingestion pipeline (issues #119 umbrella):

• X Fetcher (twscrape + cookie jar): scrape tweets, threads, media with date-range scoping (#121)
• NotebookLM helpers: ensure_notebook.py (resolve-or-create) + add_source.py (headless upload with attachments) (#126, #127)
• Dispatcher + config schema: config/ingest_sources.json + execution/ingest_dispatcher.py (run/run-one/dry-run/validate, --parallel) (#128)
• Cron scheduler: idempotent crontab install, log retention, cross-agent failure capture (#129)
• Directive + workflow + evals: directives/sources_to_notebooklm.md, .agent/workflows/sync-sources-to-notebooklm.md (#131)

Honest 1.7.10 completion:

• TurboQuant inline KV compression — finally wires the compressor into TurboQuantDynamicCache.update(). Sliding fp16 window + per-layer compressed chunks. Measured ~1.8x compression with <0.03 MSE over 64-token streams.

Deploy + safety:

• Publish gatekeeper — publish_to_public.py now uses git cherry patch-equivalence to refuse blind overwrites of public-side commits (Dependabot, CodeQL, community PRs). 8 ahead-commits correctly classified safe at deploy time.
• Airgap folder consolidation — collapsed 3 stray sibling clones to single public_release_repo/. Dockerfile + scanner skip list + .gitignore aligned.
• Worktree hook propagation — .claude/hooks/local_router_hook.py now tracked so parallel sub-agent worktrees don't deadlock.

Note on 1.7.10

1.7.10 shipped TurboQuant as a kernel + stub (compressor was real, but TurboQuantDynamicCache.update() was pass-through). 1.7.11 is the actual completion. 1.7.10 has been unpublished from NPM.

Full changelog

See CHANGELOG.md.

v1.7.9

First public airgap sync and release gate strict adherence. Suppressed false positives and officially exported securely.

v1.7.5

AGI Swarm Memory Stabilization: resolved metadata loss, BM25 text indexing bugs, fixed multi-index contradiction parsing, replaced deprecated utcnow() calls, enforced release gate, and completed full E2E testing.

v1.7.3 — Publish Workflow Fix

Patch fix: publish job now uses Node 24 (for native npm 11.x OIDC support). Previous v1.7.2 only fixed publish-github. All v1.7.1 features included (local routing, dependency tracker, security fixes, NotebookLM, system checkup, Claude dispatch, Langfuse harness).

v1.7.2 — NPM OIDC Publish Fix

Patch release to fix the NPM publish workflow. Upgrades to Node 24 (ships with npm 11.x for OIDC trusted publishing). Adds skip guard on GitHub Packages job to allow safe re-runs. All v1.7.1 features included (local routing, dependency tracker, security fixes, NotebookLM, system checkup, Claude dispatch, Langfuse harness).

v1.7.1 — Local Routing, Dependency Tracker, Security Fixes

Local-first task router with security-sensitive auto-routing to Ollama. Dependency vulnerability tracker (scans npm/pip for CVEs). Security fixes: axios→^1.8.2 (CVE-2025-27152), express→^4.21.2 (CVE-2024-29041), requests→>=2.32.0 (CVE-2024-35195), picomatch vulnerable transitive dep removed. NotebookLM skill integration. Claude Code native dispatch adapter. System checkup script (35 integrity checks). Langfuse observability harness. Security scanner false positive fixes (3 issues that blocked every release since v1.4.0). NPM OIDC trusted publisher. 6 GitHub issues closed.

v1.7.0

v1.7.0 — Parallel Agent Worktrees, Security Team Gate, Cloud Automation

Highlights

• Parallel Agent Isolation via Git Worktrees — Sub-agents can now run in parallel in isolated worktrees, preventing file conflicts. Full lifecycle: create, merge, cleanup.
• Security Team (Mandatory Pre-Release Gate) — New security_team agent team with 3 sub-agents (secret-scanner, dependency-auditor, code-security-reviewer) that must all pass before any public release. Includes Shannon entropy analysis, npm audit, license compliance, and OWASP Top 10 static analysis.
• Cloud Automation — 4-tier automation SOP: Local Agent, Claude Cowork, Cloud Tasks, Channels (Telegram/Discord).
• Session Close Protocol — End-of-session wrapup script with memory verification and cross-agent broadcasting.
• Skill Evaluation — Automated structural evaluation with Qdrant storage for historical tracking.
• Progressive Disclosure — SKILL.md files must stay under 200 lines; overflow goes to references/.

Security

The release gate now runs a full security scan (execution/security_scan.py) covering:

• 16 secret patterns + Shannon entropy analysis
• npm audit + license compliance checks
• 16 OWASP code patterns across Python and JavaScript (with CWE tags)
• Any critical/high finding blocks the release

See CHANGELOG.md for full details.

v1.6.5 — Karpathy Loop: Autonomous Skill Self-Improvement

See CHANGELOG.md for details

v1.6.4 — CI Pipeline Fix

What's Fixed

• CI publish pipeline — release_gate.py no longer hangs on interactive prompts in GitHub Actions. All input() calls are now CI-aware: auto-fail with a clear error message in non-interactive environments.
• README — 10 platforms (Kiro added), MCP compatibility documented, roadmap corrected to reflect shipped features.
• CHANGELOG — v1.6.4 entry added (was missing, causing the previous publish attempt to prompt and hang).

v1.6.3 — MCP Compatibility Layer + Kiro Support

See CHANGELOG.md for details