← ClaudeAtlas

auditlisted

Run the Plumbline engine over a repository and narrate the traceability scorecard — coverage, uncovered requirements, dead-ends, zombie code, broken anchors, orphan code-areas and warnings — prioritising the weakest areas. Use when someone asks to audit, score, or check requirement↔code traceability, or how well a codebase is anchored.
dr-zog/plumbline · ★ 0 · AI & Automation · score 68
Install: claude install-skill dr-zog/plumbline
<!-- [impl->component~audit-skill~1] --> # Plumbline — audit Run the engine, read its JSON, and turn the numbers into a prioritised scorecard. This skill only ever *reports*. Placing anchors and editing the register belongs to `onboard` and `maintain`. ## Steps 1. **Locate the engine.** Run it via the plugin's launcher, `${CLAUDE_PLUGIN_ROOT}/bin/plumbline` (it picks the right per-platform binary). Working from the engine's source tree instead? `go build -o /tmp/plumbline ./cmd/plumbline`. If neither is possible, tell the user the engine isn't available and stop. 2. **Find the register.** Default `register.md` at the repo root; pass `-register <path>` if it lives elsewhere. If there is no register at all, the repo hasn't been onboarded — say so and point at `onboard` (a later skill). 3. **Run for JSON:** ``` plumbline -json -register <register> <paths...> ``` Exit code `0` = clean, `1` = gaps found, `2` = error (bad register path, unreadable tree). Read stdout as JSON regardless of exit code `0`/`1`. 4. **Narrate the scorecard.** Lead with the `summary`: the two scores — `coveragePct` (direct needs met) and `completenessPct` (the whole chain resolves), **both computed over the approved set** — and the gap counts. Note the `gate`: `minCoverage` of 0 is strict (any gap fails); a positive value is a threshold floor; and a **spec-debt budget** (`maxProposed` count / `maxProposedPct` ratio, `-1` = no limit) fails when un-built