ada-project-audit

Run a source audit that reads every file via parallel subagents, then cross-check each "Not Implemented" requirement against actual code. Diff SRS requirement IDs against the traceability matrix with sort -u and comm to find orphans, and apply corrections before drafting any implementation plan.

Treat source and executable evidence as the source of truth for current state, then check each documentation claim against it. Verify README feature claims, appendix file lists with find output, and component line counts with wc -l rather than trusting status documents.

Use a whole-project audit for release readiness, post-refactoring verification, or legacy codebase assessment. For narrow validation of a single implementation plan before coding starts, a pre-implementation audit is the appropriate smaller-scope tool.

Legacy health mode reads docs, manifests, dependencies, and source in dependency order, then builds and tests every owned project. It cross-checks demo surfaces, JS/native call chains, TODO/FIXME markers, and stub code, ranking exploitable or data-loss defects first.

SRS documents repeat requirement IDs in cross-references, appendices, and deleted lists, so naive grep counts inflate totals. Deduplicate with sort -u and diff the ID sets; new requirements added after the matrix was generated will also be missing from traceability.

Skip it for single-file changes, bug fixes that do not touch traceability, and greenfield projects with no secondary documentation. The audit's value comes from cross-checking existing documents against code, which requires those documents to exist.