| name | audit |
| description | dure audit — run the deterministic repo scanner and surface findings (debt / missing-test / progress-consistency signals) as evidence for /dure:interview and /dure:plan. Report-only; never creates issues. |
| argument-hint | |
| allowed-tools | Bash Read |
| disable-model-invocation | true |
/dure:audit — Repo Audit
Goal: per specs/dure-audit.md, run the deterministic audit and
present its findings as evidence for planning. The interview/report runs in the user's language;
artifacts stay English (ADR-0007).
Boundary (MUST hold)
- Report-only. This skill MUST NOT write to
.dure/roadmap/ and MUST NOT create issues.
Findings are evidence; the user (or /dure:interview / /dure:plan) decides what becomes work.
- Disjoint from
dure-doctor: doctor owns integrity (schema/hierarchy/orphans); audit owns
debt / missing-test / progress-consistency signals.
Procedure
- Run
${CLAUDE_PLUGIN_ROOT}/scripts/dure-context.sh to confirm the target repo.
- Run
${CLAUDE_PLUGIN_ROOT}/scripts/dure-audit.py
→ JSON {status, counts, findings:[{check, severity, file?, line?, id?, message}]}.
- Present the findings in the user's language, grouped by
check then severity
(error > warning > info), each with its file:line or id. Lead with the counts summary.
If findings is empty, say so plainly.
- Offer next actions without auto-acting:
- feed a selected finding into
/dure:interview "<finding>" as grounding evidence,
- run
${CLAUDE_PLUGIN_ROOT}/scripts/dure-audit.py --suggest to get draft issue stubs
(title + suggested acceptance) for warning-or-higher findings, which the user can hand to /dure:plan, or
- note findings to fold into
/dure:plan.
Let the user choose; do not create issues or edit the roadmap.
Implementation status: wraps the tested scripts/dure-audit.py (3 checks: todo-marker,
untested-script, done-parent-undone-child). Auto-running on SessionStart or feeding findings
straight into /dure:plan are possible later increments, kept out of this report-only slice.
