claude-code-governance

Meta-Skill for auditing context-file hygiene. Two modes — Lite (~6.5K tokens, runs at session/task start) and Full (~150K tokens, runs on demand or weekly cron). Detects dual handoffs, version mismatches, broken cross-references, stale files, and contradictions. Master Plan §6.

Initialize Context Governance structure in a new project. Scaffolds all canonical files (CONTEXT-MANIFEST, PLAN, HANDOFF, MEMORY, CONVENTIONS, SCHEMAS-INDEX, GOTCHAS, OPEN-PROBLEMS) + CLAUDE.md governance section. Run once per project.

DEPRECATED — merged into /full-finish. Use /full-finish instead.

Universal post-task pipeline — multi-agent orchestrated audit, session handoff, docs, commit, build, release, deploy

Two-way WhatsApp communication — send progress updates and receive user replies via CC-Agent group

Open a GitHub Pull Request for review by a responsible agent (e.g., CODEX), wait for its response (text comment or 👍 reaction), remind it after 6 minutes with an @mention, and merge on approval — or after a 4-minute grace if no response. Use when the user asks to "send via PR for review", "open a PR and wait for <agent>", or to ship a reviewed change through the responsible-agent gate.

Start WhatsApp-to-Claude-Code bridge — polls CC-Agent GROUP every 60s for @cc/קלוד messages, processes as Claude Code tasks, replies in group

Live State Orchestrator. Updates PLAN.md, MEMORY.md, OPEN-PROBLEMS.md, and HANDOFF.md after milestones. Manages handoff lifecycle (active → consumed → archived). Runs after each meaningful step. Master Plan §7.2.

Plan & Execute — spawn CTO/Architect/Coder/QA/PM sub-agents to create or rewrite a comprehensive plan, review it, then implement it fully. Asks 3 intake questions to fill placeholders before running.

Pre-close reality check — scans for parallel/recent session outputs before any session close, handoff write, or release. Prevents the "partial/wrong info at close" scenario by verifying git log, file mtimes, version consistency, and active handoff count BEFORE the close skill commits to state.

Project Context Bootstrapper. Loads ONLY the relevant context for the current task using Selective Context Loading. Returns a focused briefing — current objective, current state, allowed scope, no-go zones, next safe action. Master Plan §7.1.

Enable Remote Control so the current project's Claude Code sessions can be accessed from the Android/iOS app or any browser via claude.ai/code.

Evidence Root-Cause Debugger. Takes evidence (logs, errors, symptoms) and finds the root cause with explicit confidence level. Builds a minimal fix path. Maintains separation between proven and unproven claims. Master Plan §7.4.

Impact-Safe Executor. Runs BEFORE every code write. Builds an impact map (dependencies, contracts at risk, regression zones), enforces approved scope, performs minimal edit, collects external evidence. Stops for approval if scope exceeded. Master Plan §7.3.

Orchestrate a team of specialized code agents (R&D, Product, QA, Architecture, Security, etc.) working in synergy to implement user requirements — based on Claude Code Agent Teams

Parallel Session Merge & Handoff. Reconciles outputs from multiple parallel sessions (Claude/GPT/other agents). Detects overlaps, conflicts, and contradictions. Produces a unified merged state and a fresh handoff. Master Plan §7.5.

Full QA + Security audit suite — spawns parallel QA and Security agents to verify all code changes