Tune Skill — Development skill for Claude Code
Agent skill that audits and fine-tunes a skill that already exists: triggering descriptions, sibling overlaps, stale names and counts, handoff contracts, minimal evidenced edits.
How to install Tune Skill
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open arm-pz/tune-skill and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Tune Skill does
Agent skill that audits and fine-tunes a skill that already exists: triggering descriptions, sibling overlaps, stale names and counts, handoff contracts, minimal evidenced edits. Ships check_skill.py, a deterministic SKILL.md validator. Qoder, Claude Code.
Alternatives in Development
- Tutti — Where people and agents build in tune 3.5k ★
- Magnitude — Open source inference server your agent sets up for you 1.5k ★
- Directional Prompting — Outcome-first plus directional language 138 ★
README
tune-skill
An agent skill that audits and fine-tunes a skill that **already exists**. Give it a `SKILL.md` and its siblings, and it checks the description that does the triggering, finds where two skills claim the same request, catches stale names and counts, fixes handoff contracts and command routing, makes the smallest evidenced edits, and validates them with a script rather than by eye.
It is explicitly not a rewrite. Every finding has to carry the quoted line, a severity, and the smallest fix — and the skill is required to say which checks ran, which test cases were only written, and which files it never got to see.
Install
npx skills add arm-pz/tune-skill --skill tune-skill
Or copy the `tune-skill/` folder into `~/.qoder/skills/` (Qoder) or `~/.claude/skills/` (Claude Code), then start a new session.
The validator is plain Python 3 and needs PyYAML:
pip install pyyaml
python scripts/check_skill.py /path/to/any-skill-folder
You can use `check_skill.py` on its own, without the skill, as a structural linter for any `SKILL.md`.
Ask it things like
- "This skill fires on the wrong requests — what's wrong with its description?"
- "I have two skills that both claim SEO audits. Split them cleanly."
- "Review this SKILL.md I just pasted and fix it minimally."
- "Make sure this skill hands off to its sibling instead of duplicating it."
What the validator checks
| Check | Level |
|---|---|
name is lowercase-hyphen and 64 characters or fewer |
FAIL |
| Description within 1024 characters | FAIL |
| Description has trigger wording and a boundary clause | WARN |
Folder name, frontmatter name and H1 agree |
WARN |
Every backticked references/, scripts/, examples/, assets/ path exists |
WARN |
| Body under about 500 lines | WARN |
| A stated count ("three steps") matches what follows | WARN |
Exit code 1 on any FAIL. Description is reported in characters and the body in lines, so a run is reproducible.
Layout
``
Related Skills
Few Shot
不靠 fine-tune 也能教会模型新任务——在 prompt 里塞几个示例就行。但选示例、排示例本身就是一门学问。
Owners Rep Claude Skills
Auto-triggering Claude Code skills for construction Owner's Representative work — RFIs, submittals, change ord
Plugin Tune Triggers
Sharpen a plugin's skill trigger descriptions so each fires exactly when it should
Voice Guard
Use to check ONE already-written piece of copy against a voice guide before it ships — it spawns ONE fresh sib
Whatsapp Skill
Claude Code skill that sends WhatsApp messages and searches history via steipete/wacli. Ships persistent YAML
Optimize Routing
Scan all installed plugins, detect overlaps, and regenerate ~/.claude/CLAUDE.md as an optimal routing guide. R
Related Agents
Dataset Explorer
Profiles the GeBIZ procurement data — schema, row counts, nulls, value distributions, formats, grain/keys, sup
Paper Accuracy
Review academic papers for factual accuracy. Verifies all numbers, statistics, counts, and quantitative claims
Mb Context Doctor
Diagnoses CLAUDE.md and .claude/ bloat. Runs token counts, identifies what's redundant or stale, and proposes