Audit Continue

You are resuming an in-progress audit.

Auto-Detection

1. Look for .audit-state.yaml files in audits/*/
2. Find the most recent incomplete audit
3. If multiple, ask which to continue

Resume Flow

Resuming audit: [project-name] Started: [date] Progress: X/Y items (Z%)

Last completed: [ITEM-ID] - [Title] Next up: [ITEM-ID] - [Title]

Ready to continue? (y/n)

State Recovery

Read from .audit-state.yaml:

• phase (auto-check / interactive / complete)
• Flow preference
• Remaining items

If phase is auto-check (interrupted parallel run)

⚠ Previous auto-check was interrupted. Checking which items completed before interruption...

1. Diff items_remaining against existing result files on disk (audits/<project>/<date>/<ITEM-ID>.md)
2. Items WITH result files → completed (remove from items_remaining)
3. Items WITHOUT result files → need re-checking
4. Update items_completed count from result file count
5. Re-launch parallel auto-check on truly remaining items (proceeds to auto-check phase below)

If phase is interactive

Resuming interactive review... Last item: [current_item] Remaining: [count] items needing review

Continue from current_item in interactive workflow (skip to Interactive Item Workflow below).

If phase is complete

This audit is already complete. Start a new audit with /audit-start

---

Continue using the same flow as when started.

Autonomous Evidence Gathering

Before starting the first item, read the project config (projects/<name>.yaml) and clone the repo:

CLONE_DIR="/tmp/audit-$(date +%s)"
git clone [email protected]:<owner>/<repo>.git "$CLONE_DIR"
# Fall back to HTTPS if SSH fails

Reuse this clone for all items. Do NOT ask the user for evidence you can gather yourself.

Parallel Auto-Check on Remaining Items

Before walking through items interactively, run the parallel auto-check phase on remaining items:

1. Diff items_remaining against existing result files — skip items that already have result files on disk (audits/<project>/<date>/<ITEM-ID>.md)
2. Group the truly remaining items by section
3. Launch parallel subagents (up to 8 concurrent) — one per section, using the same subagent prompt template as /audit-start
4. Present batch summary of auto-check results
5. User reviews — accept all, review failures, or drill into specifics

Interactive Item Workflow

For items marked needs-review by subagents (or items the user wants to revisit):

1. Present the item - Show ID, title, severity, section, description
2. Show the guide - Extract from checklist/checklist/[section]/guide.md
3. Run auto-checks - Run checks against the clone, don't ask the user for evidence
4. Ask follow-up questions - Only if you genuinely cannot determine the answer from the codebase
5. Determine status - Pass/Fail/Partial/Skip/Not Applicable/Blocked
6. Capture notes - Optional user notes
7. Write result file - Create audits/[project]/[date]/[ITEM-ID].md per checklist/schema/audit-result.schema.yaml (item_id not id, lowercase status, always include ## Summary, required headings per status)
8. Validate result file - Run npx tsx checklist/schema/validate.ts <result-file-path> --fix and fix any errors before continuing
9. Update state - Update .audit-state.yaml and move to next item

Regression Awareness

When auditing items that passed in a previous audit, note:

This item passed in your last audit ([date]). Let's verify it still passes.

If No Audit Found

No in-progress audit found.

Start a new audit with /audit-start <project>

Available projects:

• [list from projects/]