review-queue-auto-drain-fidelity-gate

Summary

The design record for itd-53 as re-scoped on 2026-09-21: abcd intent audit --owed, one bounded command that runs the owed fidelity audits oldest first through the host pass a single audit uses.

Scope

1. The listing comes from itd-2609150819445595's owed-intents read (receipt ids per shipped intent); --owed orders it oldest first and applies --max <n> (criterion 1).
2. Each audit: intent audit <itd-N> writes the request; the plugin page hands it to the auditor agent one at a time; intent audit ingest applies the verdict; the loop continues with the next (criteria 1, 3).
3. No reviewer: the page detects no auditor available (the host's agent listing, or a refused launch) and the command leaves every entry owed with the reason in its summary (criterion 2).
4. Not-met on a long-shipped intent is captured through capture with the receipt named, never fixed by this command (criterion 3).
5. The close hook is untouched (criterion 4).

Out of scope

• A boundary-hooked autodrain, a blocking gate, a scheduled run.
• Fixing anything: the audit-driven fix round is itd-50's, inside build.

Approach

A thin loop in the plugin page over the existing request/ingest pair, with the ordering and the cap computed by internal/core/intent from the owed list; the CLI form prints the ordered list and the next request path, so a host without the page can drive it by hand.

How the criteria are satisfied