skill-bench/Skill Eval

Evaluate Claude Code skills against YAML test cases with automated grading and PR reporting

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
skill-nameName of the skill to evaluateyes
skill-pathPath to the skill directory (containing SKILL.md and evals/)yes
anthropic-api-keyAnthropic API key for claude CLIyes
pass-thresholdMinimum pass rate (0-100) to succeedno80
timeoutTimeout per eval case in secondsno120
allowed-toolsComma/space-separated tool allow-list granted to the skill under test (e.g. "Bash(kubectl get:*),Bash(gh api:*),Read"). Forwarded to claude --allowedTools. Per-case allowed_tools overrides this.no""
permission-modePermission mode for the skill under test (default, acceptEdits, plan, bypassPermissions). Forwarded to claude --permission-mode. Per-case permission_mode overrides this.no""
post-commentPost results as a PR commentnotrue
github-tokenGitHub token for posting PR commentsno${{ github.token }}
upload-viewerUpload eval-viewer HTML as an artifactnotrue
node-versionNode.js version for claude CLI installationno22
max-retriesMax retry attempts per API call on timeout/errorno3
retry-delayBase delay between retries in seconds (multiplied by attempt number)no10
namedescription
pass-rateOverall pass rate as percentage (0-100)
passedTotal criteria passed
totalTotal criteria evaluated
cases-runNumber of eval cases executed