agentleFS
Sign inStart free
petecurranpetecurran/sorted/.claude/skills/eval/SKILL.md

eval

Skill0 starsChanged 45 days agoMITView on GitHub

Measure a photo prompt with the local model, score it against the labels, and compare it with the prompt that ships.

41 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use docs-ci, docs-interfaces and 10 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.
  • jh941213Skill

    docs-ci

    문서 CI(링크·OpenAPI lint·breaking 게이트·mermaid 검증)나 문서 드리프트 감지(docs.yaml 신선도 검사)를 설치하거나 돌릴 때 사용. 문서 내용 작성은 docs-architecture·docs-interfaces·docs-manuals.

    Also in 3 projects that use this

  • jh941213Skill

    docs-interfaces

    API·이벤트 인터페이스 문서(OpenAPI·swagger·AsyncAPI 명세, 엔드포인트 레퍼런스, API 구성도·시퀀스, API CHANGELOG)를 만들거나 코드와 맞출 때 사용. 시스템 구성도는 docs-architecture, API 설계 원칙은 api-design-principles.

    Also in 3 projects that use this

  • open-coder-aiSkill

    policy-init

    Create conformant Chock policy from request. args(request, target_path,

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-export

    Export instincts from project/global scope to a file

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-import

    Import instincts from file or URL into project/global scope

    Also in 3 projects that use this

  • diego-tobalinaSkill

    build-fix

    Fix build and compilation errors systematically. Run when build fails.

    Also in 6 projects that use this

  • +4 more from jh941213

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 32 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    Read evals/CLAUDE.md first: it has the rules.

    line 9
  2. Very specific:

    A new prompt is a new file: Write it as the next evals/prompts/vN.txt, holding exactly the text the model should see.

    line 13

“Common” is measured: kept by at least 6 in 10 of the 32 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

Prompt to measure: $ARGUMENTS (if empty, the prompt in classifier.py, which is the newest file in evals/prompts/).

Read evals/CLAUDE.md first: it has the rules. The short version is to tune on the dev set, look at the holdout once per candidate you'd ship, and never change a label to move a score.

  1. Check this machine can run the model. It needs Apple silicon (uname -sm gives Darwin arm64) and about 8 GB free. If it can't, stop and offer to re-score the existing runs instead, which needs no model.
  2. Free the memory. Two copies of the 12B model don't fit in 16 GB. If curl -s http://localhost:8800/api/health shows the app running with "mode": "gemma", ask the user before stopping it, or restart it with FT_CLASSIFIER=cache scripts/start_server.sh.
  3. A new prompt is a new file. Write it as the next evals/prompts/vN.txt, holding exactly the text the model should see. Past versions are fixed, because their runs record a hash of them.
  4. Run it in the background, since it takes about 14 seconds a photo: uv run --extra gemma python evals/run.py evals/runs/$(date +%F)/vN.json --prompt evals/prompts/vN.txt --sets dev. Add holdout only for a candidate you'd ship. Add real --real-dir <folder> if the user has the real photos, which aren't in the repository.
  5. Score it against the shipped prompt's latest run: uv run python evals/score.py evals/runs/<date>/vN.json <the shipped prompt's run>. For what changed, use --set dev --misses and --set dev --sizes.
  6. Report the table, the photos that got better or worse, and whether it beats the shipped prompt on decisions and hazards caught as well as on size. Size alone isn't a reason to ship.
  7. Only if the user decides to ship it: copy the text into classifier.PROMPT exactly, run it on the holdout set if you haven't, update the tables in docs/EVALS.md from evals/score.py, and run uv run python tests/test_units.py, whose eval checks confirm the shipped prompt is measured on today's photos.

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