← ClaudeAtlas

ce-plan-auditlisted

Independently audit a written canonical plan for structural integrity, architecture-decision quality, repository fit, decomposition, reachability, and post-write drift. Produces evidence-backed findings and never rewrites the plan. Use before specification or automated delivery when an independent plan check is warranted.
relusion/vg-sdlc-claude-plugins · ★ 0 · Code & Development · score 70
Install: claude install-skill relusion/vg-sdlc-claude-plugins
# Plan Audit **Invocation input:** $ARGUMENTS Audit a written `docs/plans/<slug>/` in a fresh context. Run deterministic integrity checks first, then inspect the decisions and delivery cut that machines cannot judge. Produce findings, not a product or release verdict, and never modify the plan. Use this workflow when plan risk, age, hand edits, or delivery cost warrants an independent preflight. It is not a mandatory ceremony for every plan: `/core-engineering:ce-plan` already runs the same hard validation floor when it writes or revises a plan. ## Runtime Inputs - **Plan:** an exact slug registered in `docs/plans/plans.json`. When omitted, use the sole registered plan or ask the human to choose among exact slugs. - **Canonical artifacts:** `plan.json`, `feature-plan.md`, `shared-context.md`, every manifest feature file, `threat-model.md`, `interaction-contract.md`, and `architecture-selection.json`. - **Conditional artifacts:** `architecture-options.md` and the governed `architecture/` package when the recorded route produced them. - **Repository evidence:** only sources cited by the plan plus one bounded hop needed to confirm an affected surface or current convention. Missing `plan.json` or another canonical required artifact is a structural finding, not an alternate plan mode. Raw intent or an unwritten plan routes to `/core-engineering:ce-plan`. ## Execution Contract 1. **Resolve safely.** Resolve the registry entry and prove every loaded path is a re