[ai] fix(ci): support stale batch in branch-cleanup-delete.sh

[github-actions]

Canonical issue

Fixes #aw_bcstale

Problem

.github/workflows/branch-cleanup.yml exposes a stale batch option (both
via workflow_dispatch and the [run-cleanup:stale] sentinel-commit path),
but scripts/maintenance/branch-cleanup-delete.sh only implemented
safe|review and fell through to usage: ... {safe|review} with exit 2
for stale. Every historical run selecting the stale batch has failed with
this exact signature since at least 2026-08-07 (15+ occurrences), most
recently https://github.com/groupthinking/EventRelay/actions/runs/34392035516
on 2026-09-09.

Fix

• Add a STALE_BRANCHES array (empty today — docs/branch-cleanup-matrix.csv
  has no CLOSE-STALE verdict rows yet) and a stale) case in
  scripts/maintenance/branch-cleanup-delete.sh.
• Update the usage string and header comment to document the new option.
• No branches are deleted by this change; the stale batch now runs as a
  documented no-op instead of erroring, matching the workflow's existing
  safe/review behavior pattern.

Verification

• bash -n scripts/maintenance/branch-cleanup-delete.sh — syntax check passes.
• DRY_RUN=1 scripts/maintenance/branch-cleanup-delete.sh stale — exits 0
  (previously exited 2 with the usage error).
• scripts/maintenance/branch-cleanup-delete.sh bogus — still correctly
  rejects unknown batches with exit 2 and the updated usage string.
• DRY_RUN=1 scripts/maintenance/branch-cleanup-delete.sh safe — unchanged
  behavior confirmed (still lists CLOSE-SAFE branches).
• git diff origin/main -- scripts/maintenance/branch-cleanup-delete.sh —
  confirms the change is an 11-line, single-file, additive diff.

Evidence

15+ failed Branch Cleanup runs over 5 weeks share this identical exit-code-2
signature (gh run list --workflow "Branch Cleanup" --status failure).

Risk

• Risk level: low
• Failure mode: none introduced — STALE_BRANCHES is empty, so stale is a
  guaranteed no-op until a future branch audit populates it.
• Rollback: revert this commit; the workflow will resume failing on stale
  batch selection as before (not a regression from current main).

Research pattern recommendation

Agentic Workflows best fits this checkpoint: the failure is a scheduled/
dispatch-triggered GitHub Actions script bug, discoverable and fixable purely
through natural-language-programmed CI automation (this very pr-iteration-loop
workflow) without needing real-time multiplayer planning (Chopin), generic
platform-embedded LLM automation (Continuous AI), or open-ended autonomous
goal-seeking (Autoloop). The fix is small, verifiable by direct script
execution, and the evidence trail (workflow run history) is exactly the kind
of structured signal agentic workflows are built to consume and act on.

Generated by pr-iteration-loop for #1778 · copilot · auto · 98.5 AIC · ⌖ 20.8 AIC · ⊞ 12.1K · ◷

• expires on Sep 16, 2026, 7:07 PM UTC

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
v0-uvai	Ready	Preview, v0	Sep 9, 2026 7:38pm UTC

[github-actions]

Commit pushed: b075a4a

Generated by pr-iteration-loop · copilot · auto · 178.1 AIC · ⌖ 11.9 AIC · ⊞ 12.1K