Kagemusha — AI skill for Claude Code
影武者 / kagemusha: a Markdown governance kit for Claude Code and AI agents—approval queues, correction promotion, and judgment distillation.
How to install Kagemusha
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open yohey-w/kagemusha and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Kagemusha does
影武者 / kagemusha: a Markdown governance kit for Claude Code and AI agents—approval queues, correction promotion, and judgment distillation.
Alternatives in AI
- ARIS Agent Guide — For AI agents reading this repo 6.2k ★
- Open Knowledge — Beautiful, AI-native markdown IDE and LLM wiki 3.7k ★
- Loushang — AI-native agent harness for coding workflows by python: multi-model LLM orchestration, stateful sessions, tool 1.3k ★
README
kagemusha
🌐 **English** (canonical) · [🇯🇵 日本語版はこちら → README_ja.md](README_ja.md)
[](https://github.com/yohey-w/kagemusha/actions/workflows/ci.yml)
**We do not ship the content of your judgment. We ship the forms.**
**Make the AI write the approval slip.** Before any outward action (send, publish, update a canonical document), the agent writes the recipient, the content, the grounds, whether it can be undone, and where it is unsure; the human reads only the 'unsure' field and signs or stops. A button approval becomes a ritual; a slip cannot be signed unread.
kagemusha is **a set of Markdown forms plus the scripts that run them**, for the AI coding agent you already use (Claude Code, Codex, Cursor, …): it makes **inward work run on its own and anything outward wait for your approval**, and it returns the reasons you rejected something to the next run as standing rules. It is not a resident agent and not a SaaS, and **your own judgment criteria are not included**.
**Fits** you if operations that cannot be undone are part of your day and you are willing to write down why you rejected something. **Does not fit** you if you want ready-made judgment criteria, or an approval SaaS for a team.
**Works with Claude Code and Codex CLI** — one instructions file, one set of forms, one command line built for whichever you run ([what maps to what](#works-with-claude-code-and-codex-cli)).
[Fixed evidence: `evidence-v1.0.0`](https://github.com/yohey-w/kagemusha/tree/evidence-v1.0.0) · [The 10-minute demo](docs/getting-started.md#the-10-minute-demo) · [Install it](docs/getting-started.md#the-steps-copy-paste)
git clone https://github.com/yohey-w/kagemusha.git
cd kagemusha
./scripts/demo-distillation.sh # ~10 min · no API key · nothing of yours is touched
*Section numbers are inheri
Related Skills
AI Software Factory
Self-hosted, open-source AI Software Factory: a Governance Controller that orchestrates heterogeneous coding a
Soulcraft
Give your agent a soul that survives model swaps. Identity kit for persistent AI agents: character, memory, an
Cyber Figures
赛博人物库 · 把中国互联网最有辨识度的人物蒸馏成 AI Skill / Claude Code Skill / LLM character distillation
Ebook Publishing Skill
The most comprehensive ebook & audiobook self-publishing skill for Claude Code — 11 platforms, HTML→PDF Puppet
Requirement Ledger
From vibe coding to knowing what you actually wanted. Point it at a project you already built with an AI agent
Save Progress
A Claude skill that stops your AI forgetting every correction you give it. Sorts what it learns by how long it
Related Agents
Promotion Officer
Use when authorizing transition between subscription modes (start_shadow shadow_to_assisted assisted_to_autono
Verifier Acceptance
Acceptance verification agent. Uses LLM judgment to assess spec coverage, acceptance criteria completion, gove
Validation Subagent
Validates generated workflow markdown files independently. Runs mechanical checks, judgment observations, rout