Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →rohitg00/skill-contract-reviewer
Validate an Agent Skill package and choose the right instruction, capability, or lifecycle primitive before implementation.
Passed 14 of 18 test cases with the skill, and 6 of 18 without it.
and 7 more in the full report
10 fixed7 passed either way5 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
After selecting two composable primitives from task-shapes.json for my skill at `./skills/code-linter`, how detailed should the rationale for each selected primitive be in the final summary report?
| Checked | Without skill | With skill |
|---|---|---|
| The response specifies providing exactly one sentence explaining each primitive selection. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill when a workflow is about to become a reusable agent artifact.
SKILL_ROOT to the absolute directory containing this installed
SKILL.md. Do not assume the process working directory is the bundle.TARGET_ROOT to the absolute original workspace working directory and
resolve the proposed skill directory under that root.$SKILL_ROOT/references/contract.md and validate the portable
SKILL.md identity fields.$SKILL_ROOT/references/decision-model.md and separate repository
context, reusable method, external capability, lifecycle timing,
deterministic logic, and isolated delegation.python3 "$SKILL_ROOT/scripts/check_skill.py" "$TARGET_SKILL", where
TARGET_SKILL is the absolute proposed skill directory under
TARGET_ROOT.$SKILL_ROOT/assets/task-shapes.json and return the smallest composable set
of primitives.Do not claim that a runtime extension is part of the portable contract. Do not treat a valid skill as permission to run scripts or access tools.
Return the validation report, the selected primitives, and one sentence explaining each selection. Include execution evidence with the resolved script path, resolved target path, cwd, exact argv, and exit code. If the host cannot expose one of those observations, mark it unverified instead of inventing it.
test cases passed, vs 6 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.