architecture-conformance-audit

name	architecture-conformance-audit
description	Thin router for the doc-vs-code architecture conformance audit (GATE-CONFORMANCE). Routes to the mechanical conformance scan plus the architecture-refresh agent loop (architecture-conformance-auditor / architecture-auditor + fixers), which own the audit behavior natively. Use before a release, after cross-package work, or when GATE-CONFORMANCE runs.

Architecture Conformance Audit (router)

The conformance audit is owned by the agent loop, not by a prose skill chain. This router only names the two layers; every judgement lives in the agents (see enforcement-architecture.md — no skill-tree tiers).

1. Mechanical floor. This step is deliberately concrete: it is the only part of the audit that must be reproducible byte-for-byte by whoever re-runs it, so the commands and output markers are named literally rather than pointed at (per the neutrality note in harness-composition-design.md).

   • Run the conformance scan — pnpm harness:conformance (see the harness:* scripts in the root package.json for the current name). Exit 0 = conformant; capture its exit code and the JSON summary printed between CONFORMANCE_JSON_BEGIN/END verbatim.
   • Run pnpm harness:scan and capture its full output as the consistency baseline for this audit.
   • Derive the dependency edge set from the manifests — for each workspace package manifest, keep the workspace-internal entries of dependencies and peerDependencies and emit one name → [deps] line per package. That the manifests, not any document, are the ground truth is a rule owned by .agents/project-structure.md (which also owns the package listing and the dependency-direction rules the scan enforces).

   This step assigns no verdicts — it produces the input step 2 diffs the documents against.

2. Agent loop. Dispatch the architecture-refresh pipeline: architecture-conformance-auditor (doc↔code claim verdicts — HOLDS/DRIFT/VIOLATION/PHANTOM/ UNDOCUMENTED — with findings + ACTIONABLE FINDINGS: <n>) and architecture-auditor (design-quality judgement), routed to architecture-fixer / architecture-implementer until a round is clean.

3. Remediation planning. When findings need follow-up backlogs + guard recommendations, use improvement-proposal-authoring.

PASS/FAIL is decided by GATE-CONFORMANCE (.agents/rules/spec-workflow.md > GATE-CONFORMANCE, run via backlog-gate-guard): the scan exits 0 and no unresolved P0 finding remains.