agentleFS
paulnsorensenpaulnsorensen/skillz-that-grillz/.agents/skills/skillz-self-update/SKILL.md

skillz-self-update

Skill1 starsChanged 23 days agoMITView on GitHub

Refresh the cross-harness research, then audit, improve, and autoimprove this repository's published skillz skill. Use when a maintainer asks to "self-update skillz", "update the published skillz skill", "refresh the harness layout", or invokes /skillz-self-update. Do NOT use for audits of other skills or as a /skillz mode.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 0 are distinctive, and 12 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 12

    Very specific

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

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

    Do not treat this as a published /skillz mode.

    line 17
  2. Very specific:

    Read skills/skillz/references/harness-layout.md, including ## Sources and ## Rejected.

    line 27
  3. Very specific:

    Check every source in ## Sources for changes since Checked:.

    line 28
  4. Very specific:

    Run /skillz audit skills/skillz/SKILL.md, then /skillz improve skills/skillz/SKILL.md.

    line 33
  5. Very specific:

    Keep the published modes and the /skillz wedge and autoimprove boundaries unchanged.

    line 34
  6. Very specific:

    Run /skillz autoimprove skills/skillz after the improve fixes land.

    line 35
  7. Very specific:

    Follow skills/skillz/references/experiments.md for the questions, the run, the gate, and the export.

    line 36
  8. Very specific:

    Run the checkout's skills/skillz/scripts/skillz-experiment.pyz, not an installed copy.

    line 37
  9. Very specific:

    Put the expected findings in expected.

    line 41
  10. Very specific:

    Never skip the case or budget approval.

    line 42
  11. Very specific:

    When the user declines a question, skip this step and report the skip.

    line 43
  12. Very specific:

    Run just build and fix any failure.

    line 47

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

skillz self-update

Maintain skills/skillz/references/harness-layout.md and skills/skillz/SKILL.md in this repository. Do not treat this as a published /skillz mode.

  1. Resolve the canonical source. Run realpath "$(git rev-parse --show-toplevel)/skills/skillz/SKILL.md". Refuse to continue when the directory is not a git checkout. Print the real path. Refuse when the real path is outside the toplevel. Confirm the checkout is this repository: git remote get-url origin matches paulnsorensen/skillz-that-grillz. Refuse a real path under ~/.agents/skills, ~/.claude/skills, ~/.omp/agent/skills, ~/.codex/skills, ~/.cursor/skills, or ~/.copilot/skills. Refuse a real path in a dotfiles-managed or npx skills-managed tree. Continue on such a copy only when the user selects it explicitly after seeing the real path.
  2. Read skills/skillz/references/harness-layout.md, including ## Sources and ## Rejected.
  3. Check every source in ## Sources for changes since Checked:. Use a fresh-context research agent when the host offers one. Cover frontmatter fields, discovery paths, invocation policy, argument handling, and body budgets. Cover Claude Code, the Agent Skills spec, Codex, OMP, Pi, Zed, and the skills CLI.
  4. Compare the research digest with the reference. Update its matrix, rules, template, and Checked: date. Record each rejected claim and its reason under ## Rejected.
  5. Run /skillz audit skills/skillz/SKILL.md, then /skillz improve skills/skillz/SKILL.md. Keep the published modes and the /skillz wedge and autoimprove boundaries unchanged.
  6. Run /skillz autoimprove skills/skillz after the improve fixes land. Follow skills/skillz/references/experiments.md for the questions, the run, the gate, and the export. Run the checkout's skills/skillz/scripts/skillz-experiment.pyz, not an installed copy. main tracks no archive, so run just build first to build it. Pass --edit prose+cli so the search can also change the inspect-skill sources in packages/skillz/src/skillz_inspect. Draft task cases that each audit one skill fixture named fixture.md. Put the expected findings in expected. The shipped contract judges them with its task rubric. Never skip the case or budget approval. When the user declines a question, skip this step and report the skip. Apply the patch only on a promote verdict and the user's approval. Then run the step 5 audit again. On a promote-cheaper verdict, do not apply the patch. Report the verdict, the delta, the SE, the token change, and the max_quality_loss margin.
  7. This repository has no deploy step. Do not install or sync the result to any installed copy.
  8. Run just build and fix any failure.

Done means: Checked: is today, every source was queried, and the published skill passes its own rubric. Autoimprove also reports a gate verdict or a recorded skip. Report the research delta, applied fixes, rejected claims, the autoimprove verdict, and the verification result.

What this skill never does

  • It never runs npx or skills add against the working copy.
  • It never applies an autoimprove patch without a promote verdict and the user's approval. A promote-cheaper verdict is advisory only.
  • It does not upgrade GEPA, skillz-experiment.pyz, the analytics engine snapshot, Python dependencies, skills/skillz/scripts/wedge.pyz, or releases.
  • It does not sync the result to other harnesses. That is out of scope for this skill.

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.