agentleFS
1017toa1017toa/agent-coach/skills/coach/SKILL.md

coach

Skill0 starsChanged 3 days agoMITView on GitHub

Grade how the user prompts Claude Code and Codex, using their own local conversation history (~/.claude, ~/.codex) and a rubric built from Anthropic's and OpenAI's official guides, then coach them in chat with rewrites of prompts they actually wrote, missed responses, and rules worth adding to CLAUDE.md / AGENTS.md. Use when the user asks to review, grade, score or improve their prompting habits, asks how well they use Claude Code or Codex, or asks for their agent-coach report or progress.

27 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use research, review and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • NVIDIASkill

    research

    Before designing a feature, research how comparable libraries solve the problem, check best practices, known pitfalls, and upstream changes. Use when starting a new feature, investigating an unfamiliar area, or the user asks to research something.

    Also in 4 projects that use this

  • SentrySkill

    review

    Fixture review skill.

    Also in 4 projects that use this

  • VercelSkill

    setup

    Set up Vercel CLI and project configuration. Use when the user says "set up Vercel", "configure Vercel", "link to Vercel", or "vercel init".

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 23 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use the first that works (check with --version):

    line 15
  2. Distinctive:

    If the user already asked plainly to grade, go on.

    line 27
  3. Very specific:

    Run agent-coach run --no-open plus the user's options.

    line 31
  4. Very specific:

    If it reports that nothing new was found, offer --days 30 or --all.

    line 33
  5. Very specific:

    Run agent-coach summary (add --lang).

    line 37

“Common” is measured: kept by at least 6 in 10 of the 23 edited versions of this file we could read. “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

agent-coach

You run the agent-coach CLI for the user and turn its output into coaching. Answer in the user's language and pass --lang ko for Korean, --lang en otherwise.

Arguments the user gave: $ARGUMENTS

1. Find the command

Use the first that works (check with --version):

  1. agent-coach on PATH (installed with pip, pipx or uv).
  2. uvx --from agent-coach-cli agent-coach or pipx run --spec agent-coach-cli agent-coach, if uv or pipx exists (the PyPI package is agent-coach-cli; the command is agent-coach).
  3. From this plugin's own files, no install needed. The plugin root holds the agentcoach package; it is ${CLAUDE_PLUGIN_ROOT}, or, if that placeholder was not filled in, the directory two levels above this SKILL.md. Run python -m agentcoach (or python3, py -3 on Windows) with PYTHONPATH set to the plugin root, e.g. PYTHONPATH="<plugin root>" python3 -m agentcoach --version. Needs Python 3.10+, no other dependencies.

Below, agent-coach means whichever form worked.

2. Preview

Run agent-coach scan with the user's window options (--days, --all, --source). Tell the user in one or two lines how many conversations will be graded (at most --max-sessions, default 20), and that each is one judge call through their own signed-in claude or codex CLI, where masked prompts and one-line action summaries are sent. Offline grading is --judge heuristic.

If the user already asked plainly to grade, go on. Otherwise ask once before grading.

3. Grade

Run agent-coach run --no-open plus the user's options. It takes about a minute per conversation, three at a time, so run it in the background if you can and tell the user it has started. Don't re-run it while it is going.

If it reports that nothing new was found, offer --days 30 or --all.

4. Coach

Run agent-coach summary (add --lang). From its output, tell the user:

  • The overall score and the three weakest habits, one line each.
  • Two or three rewrites: what they wrote, the better version, and why. Keep their wording; don't invent examples.
  • Missed responses caused by a gap in their prompt, and the "one missing piece" for each.
  • Rules worth adding, with where each belongs (CLAUDE.md / AGENTS.md, memory, skill, hook). Offer to add a rule to the right file, and only edit a file after the user says yes.

Finish with the report path it printed (~/.agent-coach/report.html) and offer to open it. Suggest running this again in a week or two; only new conversations are graded and the trend builds up.

Other commands: agent-coach history (score table per run), agent-coach report (rebuild the HTML).

Privacy

Prompts in the output may contain the user's personal details. Quote only what the coaching needs, and never send the output anywhere else. To mask names, the user can list them one per line in ~/.agent-coach/redact.txt.

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.