kanso

kanso

簡素 — elimination of clutter.

Nine Claude Code skills for people who think AI-generated code is too long.

kanso audits, refactors, commits, and writes PRs with a curatorial bias: delete before you add, match the repo's voice, earn every line.

/plugin marketplace add blakecyze/kanso
/plugin install kanso

 

The skills

Skill	What it does	Invocation
kanso-principles	Standing anti-dilution rules for code. Loaded automatically.	auto
kanso-prompting	Standing rules for writing sharper prompts. Loaded by /kanso-task.	auto (via task)
kanso-task	Rewrites a rough request into a sharp prompt, then executes it inline with kanso-principles loaded.	/kanso-task [request]
kanso-audit	Reviews code, reports findings, proposes fixes; hands refactors to /kanso-refactor and applies behaviour changes in place on approval.	/kanso-audit [scope]
kanso-nuclear	Structural maintainability review. Judo over rearrangement, file sprawl, spaghetti growth, abstraction theatre, wrong-layer logic. Defaults to the whole codebase.	/kanso-nuclear [scope]
kanso-refactor	Behaviour-preserving cleanup. Never mixes refactor with behaviour change.	/kanso-refactor [scope]
kanso-commit	Atomic commits with messages that answer why.	/kanso-commit
kanso-pr	Self-contained PR descriptions drawn from commit history.	/kanso-pr
kanso-context	Prunes and curates AGENTS.md / CLAUDE.md.	/kanso-context [mode]

 

What it looks like

Before /kanso-refactor:

def get_user(user_id):
    # Fetch user by ID
    try:
        result = db.query(User).filter(User.id == user_id).first()
        if result is not None:
            return result
        else:
            return None
    except Exception as e:
        return None

After:

def get_user(user_id):
    return db.query(User).filter(User.id == user_id).first()

Five anti-patterns removed: tautological comment, filler variable, redundant is not None check, dead else, silent exception swallow.

 

/kanso-commit output:

fix(auth): reject tokens issued before password change

Previously, changing the password did not invalidate existing sessions,
allowing a compromised token to remain valid indefinitely. Reject any
token with `iat` earlier than the user's password_updated_at.

 

Install

Plugin (recommended):

/plugin marketplace add blakecyze/kanso
/plugin install kanso

Manual, personal scope:

git clone https://github.com/blakecyze/kanso ~/kanso
mkdir -p ~/.claude/skills
cp -r ~/kanso/skills/* ~/.claude/skills/

Manual, project scope:

mkdir -p .claude/skills
cp -r path/to/kanso/skills/* .claude/skills/

Skills load on next session, or immediately via the file watcher.

 

How it behaves

• Audit reports before it edits. Findings surface in the working session; fixes land only after you approve the proposal block (y/n/edit/pick).
• Audit and refactor verify their own work. Each runs the project's lint/typecheck/test command after fixes land and pastes the exit code. No silent "looks good".
• A PostToolUse hook catches regressions inline. Every edit triggers a fast linter or syntax check on the touched file (ESLint, ruff, go vet, cargo check). Sub-second on pass, loud on fail. Opt out with KANSO_VERIFY_HOOK=0.
• Everything that writes is manual-only. Audit fixes, commits, PRs, refactors, and context edits never auto-invoke.
• Context target is AGENTS.md. CLAUDE.md is only touched when the guidance is Claude-specific.
• Voice preservation over house style. If the repo is terse, kanso stays terse.

 

Contributing

A skill has to earn its place. If you can't describe its trigger in one sentence, it's a note, not a skill. See CONTRIBUTING.md.

 

License

MIT.