skill-review

Files (9)

• SKILL.md
• composes-in-graph.yaml · fixtures/
• SKILL.md · fixtures/subject/
• run.yaml · fixtures/subject/
• composed.graph · fixtures/support/
• mark.svg
• catalog.md · references/
• mark-license.md · references/
• run.yaml

SKILL.md (first lines)

# Skill Review

Evaluate one skill from bounded evidence. The default runner passes a local
package to native `runx.skill.validate`, which parses and inspects the exact
package, runs read-only or planning harnesses with isolated receipts and no
operator credentials, and forwards that evidence to the focused assessment
runner. Execute-capable targets are inspected but never run automatically.

This skill does not mutate the package, publish a report, install a package, or
manufacture missing evidence. Use `assess` directly when a caller already has a
bounded evidence pack.

## Procedure

1. Resolve a workspace- or owning-skill-relative package path and capture its
   native capability and readiness envelope. Registry or marketplace material
   must first be installed or supplied as a bounded evidence pack; this skill
   does not fetch it implicitly.
2. Run the native harness only when catalog execution is `read` or `plan`.
3. Compare the documented capability with the inspected runner and catalog
   metadata in the evidence pack.
4. Separate native harness results, provider readback, supplied assertions, and
   unverified claims.
5. Check the happy path, refusal or stop path, authority boundary, artifact or
   effect, provenance, and recovery posture appropriate to the skill type.
6. Return `needs_more_evidence` when the evidence cannot support a trust
   decision. Never upgrade a parse result or prose claim into execution proof.
7. Recommend `adopt`, `adopt_with_caveats`, `improve`, or `reject`,
   naming the evidence and blocking gaps behind the decision.

## Stop conditions

- Refuse evidence containing secrets, raw credentials, private customer data,
  private inbox content, or provider dumps.
- Do not recommend provider readiness without provider readback.
- Do not recommend publication from private previews, placeholder hosts, dead
  links, or unrelated parent domains.
- Return `needs_more_evidence` when no native inspection, harness, receipt, or
  equivalent bounded source evidence is supplied.
- Reject capabilities that cannot be bounded, audited, or assigned a truthful
  terminal state.

## Output

- `capability_profile`: bounded capability, execution shape, and claimed effect.
- `trust_assessment`: evidence tier, caveats, and unsupported claims.
- `test_matrix`: passed, failed, skipped, and still-required checks.
- `recommendation_report`: decision, rationale, blockers, and next action.

## Inputs

- `skill` (required): local package path for the default runner; any stable
  reference may label an evidence-only `assess` run.
- `evidence_pack` (optional for the default runner, required for `assess`):
  inspection, harness, receipts, docs, or source evidence. References are
  preferable to copied private bodies.
- `objective` (optional): decision the evaluation should support.
- `test_constraints` (optional): time, environment, or safety limits.

## Agent task contracts

### `skill-review-assess`

Assess only the supplied evidence for one skill. Separate declared capability, native
inspection, harness evidence, provider readback, supplied assertions, and unverified claims.
Check the happy path, stop or refusal path, authority, artifact or effect, provenance, and
recovery posture appropriate to the capability. Return needs_more_evidence when the packet
cannot support a decision. Never infer provider readiness from a parse result, local supplied
answer, or prose claim. Recommend adopt, adopt_with_caveats, improve, or reject with
concrete blockers.

From runxhq/runx, Apache-2.0.

More from runxhq/runx