agentleFS
zyphr-labszyphr-labs/bounty-harness/.claude/skills/evaluation-harness/SKILL.md

evaluation-harness

Skill1 starsChanged 3 months agoMITView on GitHub

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are 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.

  • 3

    Very specific

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

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

    Start with run-skill-test to test a specific skill against its fixtures.

    line 13
  2. Very specific:

    Run generate-matrix periodically to get a full health dashboard.

    line 14
  3. Very specific:

    Run benchmark-runner to compare performance before/after skill changes.

    line 15
  4. Distinctive:

    Every test run has timestamp, skill version, fixture version.

    line 26
  5. Distinctive:

    Keep eval fixtures and run outputs local-only and gitignored, and redact any auth headers, session cookies, or tokens that a fixture or …

    line 30

≈ “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

Evaluation Harness

Overview

Lab fixture framework for validating bug bounty skills. Each fixture is a minimal vulnerable-by-design application that a fat skill must detect. Every fixture has a positive control (must detect) and a negative control (must not fire). Skills become measurable: precision, recall, false-positive rate, evidence completeness, runtime. The quality gate for all skill improvements.

Quick Reference

  • Skill: evaluation-harness
  • Severity range: info
  • Required tools: python3, jq, docker, curl
  • Expected input files: none

Workflow Selection

  • Start with run-skill-test to test a specific skill against its fixtures.
  • Run generate-matrix periodically to get a full health dashboard.
  • Run benchmark-runner to compare performance before/after skill changes.
  • Runbooks: use runbooks/.

Available Workflows

WorkflowPurposeScript pathsPrimary outputsEvidence
run-skill-testTest a specific skill against its lab fixtures..claude/skills/evaluation-harness/scripts/run_skill_test.py$OUTDIR/recon/eval/results_manifest.json$OUTDIR/recon/eval/evidence/
generate-matrixGenerate a full evaluation matrix across all skills and fixtures..claude/skills/evaluation-harness/scripts/generate_matrix.py$OUTDIR/recon/eval/eval_matrix.json$OUTDIR/recon/eval/evidence/
benchmark-runnerRun benchmark comparing current skill performance to previous baseline..claude/skills/evaluation-harness/scripts/benchmark_runner.py$OUTDIR/recon/eval/benchmark.json$OUTDIR/recon/eval/evidence/

Evidence Required

  • Every test run has timestamp, skill version, fixture version.
  • Positive control must trigger detection.
  • Negative control must not trigger detection.
  • Evidence completeness is scored per fixture.
  • Keep eval fixtures and run outputs local-only and gitignored, and redact any auth headers, session cookies, or tokens that a fixture or skill response may embed so they are never committed.

References

  • Source of truth: skill.yaml
  • Runbooks: runbooks/
  • Inspired by OpenAI Evals, UK AISI Inspect, OWASP testing guides

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.