AI Governance MCP — open-source meta-framework (governance engine + constitution/rules-of-procedure + infrastructure)
This repository also configures its own agents. See what ai-governance-mcp tells them →
Skill Codex
Fleet-level close-out check across every worktree and branch — is all work committed, durable on a remote, landed in the default branch, and are the finished worktrees removed? Read-only; it prints commands and never runs them. Invoke when the user says "all clear", "are we clear", "close everything out", "session…
Skill Claude Code
Assess a codebase or module's structural organization and health. Maps entry points, module responsibilities, dependency directions, and layering patterns, then evaluates coupling hotspots, cohesion gaps, and growth capacity. Invoke when the user says "architecture review", "architecture overview", "assess the…
Skill Claude Code
Run a multi-pass code review on staged changes, a branch diff, or specific files. Dispatches three specialized reviewers (correctness, security, architecture) in parallel, then reconciles findings into a severity-gated report. Invoke when the user says "code review", "review my code", "review this PR", "review these…
Skill Claude CodeCodex
Run post-change validation and finish a branch safely, including concurrent worktree refresh, optimistic fast-forward publication, retry, checkpoint, PR, or discard paths. Use after changes and before pushing, merging, handing off, or cleaning up a worktree.
Skill Claude Code
Transform raw content (transcripts, articles, lectures, informal posts) into enhanced reference documents — removes noise, restructures for clarity, fills knowledge gaps with researched best practices, preserves the author's voice. Invoke when the user provides source content to enhance, clean up, or transform into a…
Skill Claude Code
Generate or update documentation for a target file, module, API, or project. Works in three phases — classify and generate a draft, verify every claim against actual code using tool-assisted verification (grep/read as oracle), and revise any failures. Auto-detects doc frameworks. Invoke when the user says "generate…
Skill Claude Code
Run DOE calibration chunks for model/effort routing (BACKLOG.
Skill Claude Code
Mine completed session transcripts for unpersisted decisions, lessons, and context. Enriches and fact-checks memory files with HITL review. Use between sessions or when memory files feel stale relative to recent work. Triggers: "dream", "mine transcripts", "memory enrichment", "harvest sessions", "what did we miss".…
Skill Claude Code
Build an eval suite for a probabilistic / LLM-powered feature whose output has no single correct answer (generation, extraction, estimation, summarization, classification over natural-language input). Defines golden cases across three categories (control / edge / capability-limit), a scored rubric with deterministic…
Skill Claude Code
Scan the current session for unpersisted decisions, constraints, and lessons, then propose memory file updates. Use when you want to capture what was discussed but not yet written to memory files. Triggers: "journal", "capture memory", "update memory", "what did we discuss", "persist session state". Does NOT activate…
Skill Claude Code
Subagent model and effort routing table. Use when dispatching subagents via the Agent tool or Workflow scripts to select the right model and effort level. Triggers: "route this subagent", "which model for", "model routing", "effort level", "subagent model". Does NOT apply to: session model selection (that's /model)…
Skill Claude Code
Apply the Purpose-Method-Outcomes (PMO) lens to a project that turns inputs into outputs — either BUILD a project's PMO charter (who it serves, the mechanisms that deliver value, the artifacts + how you verify the purpose was actually delivered) or REVIEW an existing project against that lens (trace…
Skill Claude Code
Assess whether code is ready to be safely refactored. Maps structural contracts (what grep CAN verify must not change), warns about invisible contracts (what grep CANNOT detect — ordering, invariants, state machines), checks blast radius, and gates on test coverage prerequisites. Invoke when the user says "refactor…
Skill Claude Code
Run a focused security scan — secrets detection, dependency audit, and basic auth pattern checks. Invoke when the user says "security scan", "security review", "security audit", "check for secrets", "vulnerability check", or "scan for credentials". Do NOT use for general code review (use /code-review) or comprehensive…
Skill Claude Code
Review an external SOURCE (article, paper, repo, talk, tool, framework, blog post) against ai-governance. Extracts the source's items, abstracts each to the INTENT above the surface, checks whether the framework already covers that intent, and PROPOSES where each verdict lands (INFLUENCES.md row / reference capture /…
Skill Claude Code
Onboard a project under the AI-governance framework — calibrate the mode, ask the minimal founding questions (Goal / Done-looks-like / Non-goals / app-vs-document), then lay down the governance + memory scaffolding via scaffoldproject. Invoke when the user says "start a project", "new project", "set up this project"…
Skill Claude CodeCodex
Safely start, resume, validate, or clean up an isolated Git worktree for a mutating Claude Code, Codex CLI, or Codex Desktop session. Use at the start of concurrent work, when moving a session off the primary checkout, when recovering a partially prepared worktree, or before removing a completed worktree.
Skill Claude Code
Generate comprehensive tests for a target file, function, module, or diff. Works in three phases — generate tests with error-path emphasis, verify against known AI test-generation failure modes (echo-chamber assertions, error-path undercount, weak mutation coverage), and revise any that fail. Auto-detects test…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: