Skip to content

/gz-intent-trace

Trace a sampled set of ADRs from declared intent to shipped surface, and route every gap as a correction under its owning ADR.

Purpose

/gz-intent-trace answers one question per ADR: does the shipped surface fulfil its original declared intent? Every no is a correction routed under the owning ADR — never a fresh pool ADR, never new-design ceremony, and never an "enhancement". That routing is operator canon: discovering that more is needed to fulfil the intent of a feature is a correction, not an enhancement.

Distinct from /gz-adr-audit, which asks whether an ADR's evidence chain supports closeout. This skill asks whether the thing that shipped does what was decided.

When to Use

Invoke after an inertness audit surfaces candidates, before a release, or when the operator reports that a capability "exists but doesn't really work". It is the deepest axis of /gz-health-audit and is sequenced last on purpose.

What to Expect

A sampled diagnosis with routing, not an exhaustive pass. The skill samples by risk signal — parked/withdrawn/repudiated/uncovered OBPIs, measured-inert mechanisms, heavily-cited foundation ADRs, ADRs already failing their own evaluation binding — and records the sample before reading any of it.

Output is a per-claim verdict (fulfilled / correction / enhancement) with the declared claim quoted verbatim and the observed surface pasted, plus a route for each correction.

Boundaries

  • Diagnosis and routing only; never implements the corrections it finds
  • Never widens the sample mid-run
  • Never authors a new ADR
  • Surfaces campaign collisions to the operator rather than resolving them

Invocation

Text Only
/gz-intent-trace

The canonical execution contract lives at .gzkit/skills/gz-intent-trace/SKILL.md (mirrored into .claude/skills/, .agents/skills/, .github/skills/).