What's New

LACP is now a Claude Code plugin. Same battle-tested hooks, native distribution.

Claude Code Plugin

Install with one command:

git clone https://github.com/0xNyk/lacp
claude --plugin-dir ./lacp/plugin

5 lifecycle hooks:

• Stop — quality gate with 4-dimension weighted scoring + handoff artifacts
• SessionStart — git context, focus brief, handoff injection, SMS memory
• PreToolUse — blocks rm -rf, co-author injection, publishing, exfiltration
• PostToolUse — continuous QA (test every N writes with thinking prompts)
• UserPromptSubmit — thinking nudge for deeper reasoning

3 skills + 11 context modes (tdd, debugging, sprint, verification, brainstorm, think, orchestrate...)

E2E tested: 18/18 tests pass.

Native TUI REPL

lacp (no args) launches a Textual TUI with:

• Multi-provider support (Anthropic OAuth, OpenAI, Ollama)
• 17 built-in tools, MCP integration, session persistence
• Responsive layout (40-200+ columns)
• Claude Code-style spinner + Hermes-style tool display

Unified Brain CLI

lacp brain — 16 subcommands for Obsidian vault operations:

• search, read, write, links, orphans, hubs, graph
• episodes, expand, sync, review
• REST API + ripgrep dual fallback

Capability-Gap Registry

lacp gaps — council-informed (Musashi + Sun Tzu + Torvalds) framework for deciding when to build a runtime vs stay a wrapper. Currently 3/5 hard blockers.

OAuth Fixes

Matched Claude Code's full client config after reverse-engineering free-code (TypeScript source). Billing header, User-Agent, 10 retries with exponential backoff.

Research

• Analyzed all 21 Anthropic engineering articles (10 produced code)
• Full market research: TAM, competitors, monetization strategies
• User voice: Reddit, HN, GitHub issues (45K+)
• Identified Ori-Mnemos and obra/knowledge-graph for future integration

Full diff: v0.5.0...v0.6.0

Thinking Partner System

Five-part "thinking partnership" layer for AI sessions, bridging the gap between enforcement (quality gates) and generative thinking support.

New Commands

• lacp-focus — Living focus brief with 4-question template (current problem, beliefs/uncertainties, open decisions, 30-day goal). Auto-injected at session start with staleness warnings after 7 days.
• lacp-reflect — Weekly reflection CLI pulling provenance chain, generating context-specific prompts, reporting focus brief staleness.

New Hooks

• Blind spot analysis (stop_quality_gate.py) — Opt-in Ollama step surfacing unchallenged assumptions and avoided questions before allowing stop. Enable with LACP_BLIND_SPOT_ENABLED=1.
• Thinking nudge (thinking_nudge.py) — UserPromptSubmit hook detecting bare questions without stated positions. Opt-in via LACP_THINKING_NUDGE=1 or LACP_CONTEXT_MODE=thinking-partner.

New Context Modes

• thinking-partner — Challenge-first protocol ("what am I not seeing?")
• implementation — Focused building mode
• review — Critical analysis mode

New Hook Profile

• thinking-partner — installable via lacp-claude-hooks apply-profile --profile thinking-partner

Architecture

Other Changes

• README restructured with best-practice patterns
• Agent-id regex fix for 128-bit IDs

Install / Upgrade

brew tap 0xNyk/lacp
brew install lacp
# or upgrade
brew update && brew upgrade lacp

Full Changelog: v0.3.1...v0.4.0

What's New

Security Audit Remediation (#22)

• 2 critical fixes: command injection in quality gate (shell=True → shlex.split), shell injection in runner scripts (printf %q escaping)
• 5 high/medium fixes: predictable /tmp paths moved to ~/.lacp/hooks/state/, unsafe .env sourcing replaced with strict parser, enterprise policy hardened
• 6 low fixes: session ID sanitization, full SHA-256 in provenance chain, agent ID entropy increased to 128-bit

Brain-Expand Pipeline Scripts (#23, #18)

5 missing automation scripts implemented (community contribution by @w3t-wr3, hardened with security fixes + 20 direct tests):

• run_session_sync.sh — Claude/Codex session → inbox
• detect_knowledge_gaps.py — orphans, broken links, stale notes
• generate_review_queue.py — prioritized review queue
• route_inbox.py — frontmatter-based routing to graph subdirectories
• archive_inbox.py — stale note archiving

Nightly Automation (#24)

• lacp brain-nightly — cron-ready wrapper with pre-flight checks (disk, GPU VRAM), log rotation, --install-cron
• GPU resource management: gpu_pressure_ok(), acquire_gpu_lock() / release_gpu_lock() (portable macOS fallback)
• Memory decay zones: identity (0.1x), knowledge (1.0x), operations (3.0x)

First-Class Hermes Support (#16)

• Hermes as first-class wrapper/runtime across LACP scripts
• lacp brain-promote command with CI coverage
• Stable Obsidian plugin recovery (0 drift)
• README marked stable, alpha messaging removed

Other

• Schema-gated ingest, resolver, KPI, Obsidian graph optimization (#19)
• Agent identity registry + provenance chain (#13, #14)
• Modular Python hook pipeline with test verification (#12)
• README refresh with langdag-inspired structure (#20)

Stats

• 16 PRs merged since v0.3.0
• 60+ integration tests
• 0 open issues

What's New

68 commits, 152 files changed, 18K+ lines added since v0.2.0.

🧠 Brain Stack & Knowledge Management

• brain-stack — composable brain bundles for layered knowledge contexts
• brain-ingest — bulk knowledge ingestion pipeline
• brain-doctor — health checks with auto-discover formula and remediation hints
• brain-expand — gap detection, FSRS review queue, weekly research consolidation, obsidian graph health guard
• mcp-health — deep QMD validation for MCP server health
• Repo → Obsidian mirror — repo-research-sync pipes research into brain graph
• Obsidian brain bundle — automation profiles for vault management

🔒 Security & Compliance

• security-hygiene — audit command surfacing intervention KPI in status
• Intervention rate per 100 with baseline deltas in reports
• Hardened execution hook profiles for Claude
• Quality-gate profile with ollama-backed Stop hook
• Session fingerprint gate for context drift control

🖥️ Interactive Console

• lacp console — interactive REPL with slash commands
• /loop shortcut — recurring task execution from console
• Session time tracking — auto-track console session duration
• Monthly time tracking with project-level report rollups
• Context profile packs for safer loop execution

🚀 Orchestration & Swarm

• lacp up — dmux-style multi-instance sessions with layout presets
• Swarm status — collaboration conflict summary and reservation collision analysis
• Runtime pressure guardrails for orchestrate and swarm modes
• Trace triage and plan-to-act handoff contract enforcement
• Loop control with hardened loop gates

🔧 Harness & Testing

• Harness replay for failed task receipts
• Cascading task IO contracts enforcement
• Expected output artifact validation against schemas
• Failure classification with remediation plans
• Stabilized test runner with recursion guards

📦 Release & Bootstrap

• Auto-deps bootstrap with canary clean-baseline mode
• Local publish flow with hardened wrapper adoption
• Open-source readiness scrub — removed personal paths, simplified defaults
• Auto-install obsidian and qmd in bootstrap
• Shared skills root — linked claude and codex skill trees

🛠️ Skills & Feedback

• skill-score — skills-sessions feedback loop with confidence scoring
• Anthropic skill sync command for skill-creator
• Posture command for local-first contract enforcement
• Hook audit and repair tooling
• Hook optimization profiles and onboarding auto-tune

📝 Docs

• Daily developer workflow runbook
• Obsidian brain bundle and automation profile docs
• Updated README with new command reference

Full Changelog: v0.2.0...v0.3.0

LACP v0.2.0 Release Notes (Draft)

Highlights

• Added lacp canary for 7-day promotion readiness gates.
• Added lacp auto-rollback for fail-safe fallback when canary is unhealthy.
• Added lacp schedule-health to run periodic local health checks via launchd.
• Added lacp policy-pack with starter, strict, and enterprise baselines.
• Added lacp release-prepare for one-command go/no-go discipline.
• Extended lacp report with wrapper observability metrics.
• Unified status and report JSON top-level schema fields.

New Commands

lacp canary --json | jq
lacp auto-rollback --json | jq
lacp schedule-health install --interval-min 60 --json | jq
lacp policy-pack list --json | jq
lacp release-prepare --quick --skip-cache-gate --skip-skill-audit-gate --json | jq

Validation

• Full isolated suite passed:
  • bin/lacp-test --isolated
• Added CI coverage for:
  • canary
  • auto-rollback
  • schedule-health
  • policy-pack
  • release-prepare

Upgrade Notes

• Homebrew formula command exports were expanded to include all new binaries.
• Existing install flows remain compatible (install.sh, brew, direct clone).