alibaba/Agent Skill Eval (skill-up)

Run skill-up behavioral evals on your Agent Skills in CI — across claude_code, codex, qodercli, and qwen_code engines.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
engineAgent engine: claude_code / codex / qodercli / qwen_code. Leave empty to let the repo's eval.yaml declare it.no""
modelPassed to `skill-up run --model` when non-empty; otherwise the eval config / CLI default is used.no""
providerResolves base-url from a built-in table (public: dashscope). Ignored when base-url is set.no""
api-keyModel credential. engine non-empty -> routed to ANTHROPIC_API_KEY / OPENAI_API_KEY / QODER_PERSONAL_ACCESS_TOKEN; engine empty -> passed via `skill-up --api-key`. Always source this from a GitHub secret.no""
base-urlModel API endpoint (public). Defaults to the agent CLI's own public endpoint when empty.no""
open-sandbox-api-keyOPENSANDBOX_API_KEY; only needed when evals use the opensandbox environment.no""
skill-targetSkill directory or eval config path, resolved against the checked-out repo root.no.
skill-up-versionskill-up version. Ignored when the image already ships skill-up (the default).nolatest
skill-up-commandThe skill-up command itself; must start with `skill-up`.noskill-up run
parallelismPassed to `--parallelism` when non-empty (1-256).no""
agent-install-commandOptional custom install command, for a custom image missing a target engine CLI.no""
namedescription
exit-codeskill-up exit code (0 = all evals passed).
report-dirDirectory with the JSON / JUnit / HTML reports, relative to the workspace (e.g. skill-up-workspace), so host steps can consume it.