agentleFS
Sign inStart free
TrimVisTrimVis/skills/refine-skills/SKILL.md

refine-skills

Skill2 starsChanged 3 months agoMITView on GitHub

Mine Claude Code transcripts for how your installed skills actually behaved, find recurring corrections, and propose SKILL.md edits one finding at a time. Use when the user says "refine my skills", "improve my skills from usage", "what keeps going wrong with skill X", or invokes /refine-skills.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. Silence-view/llm-language★ 2same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository8 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the output. Do not grep it for complaint words — that finds pasted test output, not complaints.

    line 28
  2. Distinctive:

    Corrections: Follow-up turns that redirect the run.

    line 39
  3. Distinctive:

    Ask to apply, skip, or edit.

    line 54
  4. Distinctive:

    Keep each edit to a few lines.

    line 56
  5. Distinctive:

    Do not read whole transcripts into context.

    line 60
  6. Distinctive:

    Do not report a finding you cannot quote an episode for.

    line 61
  7. Distinctive:

    Do not edit a skill the user did not select.

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Refine Skills

A correction the user retypes at every invocation is a default missing from the skill. This mines the transcripts for those and proposes the edit.

Step 1 — Select

python3 ~/.claude/skills/refine-skills/skill-usage.py list

Show the table. Ask which skills to analyze by number, and accept all. Do not pick for the user.

Skills under ~5 episodes rarely yield a pattern. Say so and let the user decide.

Step 2 — Extract

Per selected skill, all time:

python3 ~/.claude/skills/refine-skills/skill-usage.py extract <skill>

Read the output. Do not grep it for complaint words — that finds pasted test output, not complaints.

Step 3 — Find

A finding needs two or more episodes on different days. One annoyed message is not a pattern.

Look for:

  • Repeated args. The same instruction passed again and again (--keep-alive, use a fable agent, also /commit). The strongest signal, and the easiest to fix.
  • Chaining. Args that name another skill (and then /difit-open) — the skill should offer or do the next step.
  • Retries. RETRIED after Nm means the first run did not land.
  • Corrections. Follow-up turns that redirect the run.

Then open the SKILL.md and confirm the gap is real. If the skill already says it and was ignored, the fix is wording or placement, not a new rule.

Confirm the fix is buildable before you propose it. A rule that reads state no tool exposes is dead on arrival, so check the CLI, the API, or the file format first. When only part of a finding is mechanizable, propose that part and say which part is not.

Ignore one-off task context (push-api-2, mock-api). That is the user steering, not the skill failing.

Step 4 — Propose

One finding at a time, worst first. For each, show:

  1. The finding and the episodes that support it (dates, quoted args).
  2. A diff against the SKILL.md.

Ask to apply, skip, or edit. Apply the accepted diff, then move to the next finding. Never batch them.

Keep each edit to a few lines. These files are instructions and they get longer every pass, so replace a rule where you can instead of appending one.

Don'ts

  • Do not read whole transcripts into context. The script output is the input.
  • Do not report a finding you cannot quote an episode for.
  • Do not edit a skill the user did not select.
  • Do not commit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free