Trace declared intent to observed shipped behavior and classify gaps as corrections rather than new design. Use when a capability exists but may not fulfill its specification, during a pre-release integrity review, or when checking whether completed work actually delivered what was decided.
Resources
1Install
npx skillscat add tvproductions/gz-skills/gzs-intent-audit Install via the SkillsCat registry.
GovZero Intent Audit
Answer one question per sampled capability: does the observed surface fulfill
the intent that authorized it?
Discovery and fallback
Prefer, in order:
- The project's own traceability surface, when one links decisions to
delivered behavior. - Repository tooling that exercises the delivered surface at its user-visible
seam. - The authoritative document read directly, and the surface exercised by hand
at that same seam.
The third rung is the floor and is what the audit actually rests on. No project
tooling can substitute for reading the claim and observing the behavior, so
this skill is never blocked by a repository that offers neither.
Workflow
Resolve the audit population. For a large corpus, select a bounded sample
before reading implementation, using risk signals such as incomplete work,
stale evidence, failing checks, inert mechanisms, high fan-out, or user
reports. Record why each item entered the sample.Read the authoritative decision, specification, issue, or acceptance record.
Extract each intent claim verbatim or with an exact citation. Do not inspect
the implementation until the claims are fixed.Locate the actual delivered surface. Run or exercise it at the user-visible
seam; a test or documentation claim alone does not prove behavior exists.Compare each claim with observed behavior:
- Fulfilled: the claim is demonstrably met.
- Correction: the delivered surface does not fulfill its owning intent.
- Enhancement: the intent is fulfilled and a genuinely new capability is
proposed.
Route corrections under their owning work item or decision. Do not create a
new design artifact merely because the original delivery is incomplete.Stop at diagnosis and routing unless the user also requested implementation.
Evidence
Report the sample and selection signals, cited intent claims, commands or steps
used to observe the delivered surface, per-claim verdicts, and exact correction
routes. Name unavailable evidence and uncertainty; never infer fulfillment from
artifact presence alone.