coverage-audit

# coverage-audit — read-only cross-artifact consistency analyzer > **Activation**: manual-only via `/spawn coverage-audit <PLAN-NNN>`. > Never invoked by `/debate`, `/execute`, or `/ceo-boot` by default. > Tier-A defensibility per ADR-125 §143-150 (zero token spend in steady state). ## Purpose Port of github/spec-kit's `/speckit.analyze` (`templates/commands/analyze.md:L340-L381`). Performs 6 detection passes across the loaded PLAN + canonical roadmap + findings-master + ADR ledger. **STRICTLY READ-ONLY** — never modifies files (enforced by `read_only: true` frontmatter + skill text contract). ## 6 detection passes (severity-classified) | # | Pass | Severities flagged | |---|------|---| | 1 | Duplication | HIGH (same AC text in 2+ plans) / MEDIUM (paraphrase) | | 2 | Ambiguity | HIGH (vague terms — "should", "may", "TBD"; **LIVE `[NEEDS CLARIFICATION: …]` marker** — PLAN-SCHEMA §14) / LOW (jargon w/o anchor) | | 3 | Underspecification | CRITICAL (AC missing path) / HIGH (AC missing severity) | | 4 | Constitution Alignment | CRITICAL (CLAUDE.md drift vs PLAN-NNN.related_adrs) | | 5 | Coverage Gaps | HIGH (AC -> no test file) / MEDIUM (AC -> 1 test only) | | 6 | Inconsistency | HIGH (status conflict draft vs reviewed) / LOW (date drift) | Severity tiers: CRITICAL / HIGH / MEDIUM / LOW. Limit: **50 findings max** per invocation; overflow summarized with `+N more findings (run with --no-cap for full output)`. ### Pass #2 (Ambiguity) — inline clari...