bug(unic-archon-dlc): /pr-review's review-gate cannot re-pause, its message names an unresolved path, and prune_slug_dirs is off

[orioltf]

What this is

Three leftovers from #439's grilling on 2026-09-04, approved by the maintainer as a spin-out and separated from #439 because none of them changes what /build produces. #439's first spin-out shape ("no generated artefact is committed to a merged branch") died in that grilling: cleanup.prune_slug_dirs already removes <artifacts_dir>/<slug>/ once a slug's pull request is merged or closed, and it defaults to false, so the accumulation on develop has a built, unused fix.

1. /pr-review's review-gate has no on_reject. A reject cancels the run. The findings are on disk (synthesize and reconcile write four SESSION/* files before the gate), so a reject loses nothing until archon complete <branch> --force deletes the worktree. A human who wants to read before deciding has no re-pause, unlike /build's build-pr-gate, whose on_reject re-pauses with max_attempts: 3.
2. The gate message names SESSION/comment.md and SESSION/threads.json as paths a human should open, and SESSION is a constant the message never resolves, so the human is pointed at a path that does not exist as written.
3. Whether cleanup.prune_slug_dirs should default to true. Session artefacts stay committed by design (bug(unic-archon-dlc): open-pr precedes its gate, and a config-value gate has no safe middle #439 criterion 6 and the 30 712-character report that ruled out a comment); the mechanism that prunes them after merge exists and is off.
4. Whether that pruning wants its own gate mode. Added 2026-09-04 22:30 on the maintainer's words in bug(unic-archon-dlc): open-pr precedes its gate, and a config-value gate has no safe middle #439's grilling: "Creo que esto puede gozar de una configuracion AFK|HITL tambien, no? Y un gate para HITL, claro. Va al B." So the question is not only the default but whether prune_slug_dirs becomes a gates.<box>-shaped setting with an approval: node in hitl. Note the collision with the two-valued gate vocabulary bug(unic-archon-dlc): open-pr precedes its gate, and a config-value gate has no safe middle #439 measured (hitl | afk, no middle) before adding a third instance of it.

Open question, to be grilled before criteria

Is an on_reject that only re-pauses a gate, or a workaround for a schema with no on_approve (ADR-0011 line 43)? The maintainer's own experience in run 2 was "the only lever was destroying the run", and item 1 above is that lever.

Related

#439 (source), #458 (/qa's unread node outputs, not absorbed here), #440 (re-entry idiom).