agentleFS
Sign inStart free
gcontinigcontini/ats-strike/.claude/skills/cv-eval/SKILL.md

cv-eval

Skill5 starsChanged 11 days agoMITView on GitHub

Evaluate ats-strike CV quality on the fixed test set. It generates the CVs (quick = 10, thorough = 50), has the cv-judge agent score each one for ATS appeal, AI-written style and readability, and appends the averages to tests/cv_quality/results.csv. Use after a serious prompt or architecture change, or when asked to evaluate, benchmark or score CV quality. Arguments - quick|thorough [temperature].

  • Reads credentials

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Requirements: Every JD of the set needs tests/cv_quality/jds/<stem>.requirements.json.

    line 24
  2. Distinctive:

    Never rewrite an existing requirements file.

    line 28
  3. Distinctive:

    Errors: if it stops on a malformed judge file, run cv-judge again for that CV only, then score again.

    line 41

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Evaluate CV quality

Arguments are the set, quick (default) or thorough, and an optional temperature for the cv step. It is sent as that step's whole extra_body, so the step's other models.toml settings are dropped for the run. Run everything from the repository root.

  1. Generate. Each CV takes minutes, so run this in the background and wait for it to exit:

    uv run python tests/cv_quality/cv_eval.py generate <set> [--temperature T]
    • Server: it starts this checkout's own ats-strike on a free local port, uncommitted changes included, and stops it at the end. It needs .env with the model keys and pdflatex on the host. Its log is <run>/server.log.
    • Candidate: candidate_profile.json and candidate_data.json come from examples/candidate/, the fictional candidate in the repository, so runs stay comparable. --candidate <folder> writes for someone else; the folder is recorded in run.json.
    • Prompts: the client still uploads the prompts and template it finds in ats-strike-data/gab, as in real use, and those win over server/resources. A prompt edited only in server/resources is not tested while a copy sits in ats-strike-data/gab.
    • Run folder: the last line printed is the run folder. A failed CV leaves a <stem>.error.txt there instead of a CV.

  2. Requirements. Every JD of the set needs tests/cv_quality/jds/<stem>.requirements.json. The first ten JDs are the quick set. For each missing file, run the cv-judge agent with this message:

    Mode "requirements". JD: tests/cv_quality/jds/<stem>.txt. Write tests/cv_quality/jds/<stem>.requirements.json.

    Never rewrite an existing requirements file. It is the yardstick earlier runs were scored with.

  3. Judge. For each <stem>.cv.json in the run folder, run one cv-judge agent, five at a time, with this message:

    Mode "judge". JD: tests/cv_quality/jds/<stem>.txt. Requirements: tests/cv_quality/jds/<stem>.requirements.json. CV: <run>/<stem>.cv.json. Units: <run>/<stem>.units.txt. Catalog: tests/cv_quality/ai_tells.md. Write <run>/<stem>.judge.json.

  4. Score.

    uv run python tests/cv_quality/cv_eval.py score <run>
    • Report: give the per-CV table, the averages, the judge model it names, and the row appended to tests/cv_quality/results.csv.
    • Errors: if it stops on a malformed judge file, run cv-judge again for that CV only, then score again.

Reading the scores:

  • appeal: half ATS keyword coverage, half experience shown, weighted must = 2, nice = 1.
  • ai: the share of CV sentences that carry AI tells.
  • readability: the share of sentences that read without stumbling.

Higher appeal and readability are better, and lower ai is better. With ten CVs, small differences are noise, so decide on thorough runs.

The judge is noisy. On 2026-10-16 the same 3 CVs were judged twice, and per-CV scores moved by up to 3 points for appeal, 4 for ai and 12 for readability. Readability is the noisiest because its flags are few.

Download now (ZIP)

  • Other versions1 project uses a version of this.

    Word for word · 1

    1. gcontini/resumix★ 5100% same

    Edited · 0

    None found.

    Wrote their own version · 0

    None found.

    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 repository3 other files it gives its agents.

Does it work?

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

Reviews can't be read right now.

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