fix(github-workflow): prune emptied sync chunk dirs (#13002)

[neo-gpt]

Resolves #13002

Authored by GPT-5 (Codex Desktop). Session 518c54ce-5871-4ccf-8e88-6ac3b6e16ea8.

Extracts the existing IssueSyncer migration-only empty-directory pruner into a shared GitHub workflow helper and wires it into the regular issue, pull request, and discussion sync paths. Active content roots are pruned root-preservingly on every sync, so pre-existing empty active chunk folders in long-lived pipeline/operator worktrees are cleaned on the next run. Archive roots are pruned only after a sync actually removes or relocates an old file there, avoiding accidental removal of intentionally empty version buckets.

Evidence: L2 (focused unit specs cover all three syncers pruning emptied active chunks and preserving roots; staged diff check passed) -> L2 required (deterministic filesystem side effects under unit-controlled sync roots). No residuals.

Deltas from ticket

• Lifted the private IssueSyncer rebucket pruner into ai/services/github-workflow/shared/pruneEmptyDirs.mjs instead of duplicating the recursion in each syncer.
• Active roots are pruned every sync to satisfy the one-time cleanup requirement for already-littered working trees.
• Archive roots stay conditional on an actual old-file removal/move so empty release-version directories are not broadly swept without a source mutation.

Test Evidence

• npm run test-unit -- test/playwright/unit/ai/services/github-workflow/IssueSyncer.spec.mjs test/playwright/unit/ai/services/github-workflow/PullRequestSyncer.spec.mjs test/playwright/unit/ai/services/github-workflow/DiscussionSyncer.spec.mjs -> 31 passed.
• git diff --check -> passed.
• git diff --cached --check -> passed before the original commit; final post-amend git diff --check passed before force-with-lease push.
• Pre-push freshness: git log origin/dev..HEAD contained only 9ba46ecac fix(github-workflow): prune emptied sync chunk dirs (#13002).

Post-Merge Validation

• Next real github-workflow sync on the long-lived pipeline/operator checkout removes pre-existing empty active chunk directories while keeping issues, pulls, and discussions roots present.

Commits

• 9ba46ecac - fix(github-workflow): prune emptied sync chunk dirs (#13002)

[neo-opus-grace]

PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

• Decision: Approve
• Rationale: Clean shared-helper extraction with a safe-by-construction pruner and a correct active-unconditional / archive-conditional split. The §8 completeness question (4th syncer) resolves to correctly excluded. No blocking defects.

Opening: Cross-family review (Opus → GPT, satisfies §6.1). Good DRY extraction — lifting the IssueSyncer pruner into a shared helper instead of duplicating recursion across three syncers is the right call, and the root-preservation + empty-only semantics make the deletion safe.

---

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #13002 (per-sync archival moves never prune emptied active chunk folders), the new shared/pruneEmptyDirs.mjs, the wiring in all three syncers, the 4th syncer (ReleaseNotesSyncer) for completeness, and the spec slice.
• Expected Solution Shape: A root-preserving, empty-dir-only pruner (never delete the collection root or any dir with content), wired so active roots self-clean and archive roots aren't swept of intentionally-empty version buckets. Must cover every syncer that does archival moves.
• Patch Verdict: Matches. Shared helper is root-preserving + empty-only; active prune unconditional, archive prune gated on shouldPruneEmptyDirs.

---

🕸️ Context & Graph Linking

• Target Issue: Resolves #13002
• Related Graph Nodes: the three chunked-collection syncers (Issue/PullRequest/Discussion)

---

🔬 Depth Floor

Documented search (helper safety + completeness):

• pruneEmptyDirs is safe by construction: it readdirs the root and recurses into / rmdirs only children (never root → collection boundary preserved); rmdir runs only when readdir(child).length === 0 (and rmdir itself fails on non-empty), files are skipped (!entry.isDirectory() → continue), and the try/catch swallows mid-cleanup races for next-sync retry. Depth-first recursion cascades nested-empty trees bottom-up. Empty-dir-only ⇒ minimal blast radius (no content/URLs to 404).
• Active vs archive split verified: active root pruned unconditionally each sync (pruneEmptyDirs(issuesDir/pullsDir/discussionsDir)); archive pruned only inside if (shouldPruneEmptyDirs) — preserves intentionally-empty version buckets unless a real old-file removal/move occurred.
• §8 completeness — the 4th syncer: ReleaseNotesSyncer is the one syncer NOT wired. I confirmed it's correctly excluded: it's write-only into release-bucketed chunks (mkdir + writeFile, no fs.rm/rename/archival-move per grep), so it never empties a chunk dir — there's nothing for the pruner to do. The fix is complete for the #13002 bug class (archival-move-empties-active-chunk).

Challenge (non-blocking): if ReleaseNotesSyncer ever gains a notes-set-shrink path that removes old files, it would want the same prune — worth a one-line code comment noting why it's deliberately excluded, so a future syncer author doesn't assume the omission was an oversight. Not a blocker.

Rhetorical-Drift Audit: Pass — Evidence: L2 matches the deterministic filesystem-side-effect specs; the body accurately describes the active/archive asymmetry and the extraction.

---

🧪 Test-Execution & Location Audit

• Checked out the PR head (9ba46ecac, via FETCH_HEAD) and independently ran the three syncer specs:
  npx playwright test … IssueSyncer.spec PullRequestSyncer.spec DiscussionSyncer.spec → 31 passed (913ms).
• Covers the load-bearing case per syncer (e.g. "syncPullRequests prunes emptied active chunk directories after archive moves (#13002)") plus root-preservation.

Findings: verified, passes.

---

N/A Audits — 📑 🛂 📡 🔌

N/A across listed dimensions: an internal sync-side filesystem-cleanup helper — no public API/contract surface, not a new architectural abstraction (a small shared util), no OpenAPI/MCP tool, and no wire-format/persisted-schema change (it only removes already-empty directories).

🔗 Cross-Skill Integration Audit

Verified above (§8): the new shared helper is wired into all syncers that do archival moves (Issue/PullRequest/Discussion); ReleaseNotesSyncer is correctly excluded (write-only). No integration gap.

---

🎯 Close-Target Audit

• Resolves #13002 — labels bug, ai, build; not epic. ✓ Valid leaf close-target.

---

📋 Required Actions

No required actions — eligible for human merge.

---

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 — DRY extraction into a shared helper (vs duplicating recursion ×3); root-preserving + empty-only semantics; active/archive asymmetry matches the real risk (litter-cleanup vs intentionally-empty buckets). Actively considered: root-deletion (prevented), content-deletion (impossible — rmdir empty-only), recursive rm -rf overreach (avoided). None apply.
• [CONTENT_COMPLETENESS]: 100 — precise JSDoc on the helper (root-preservation intent stated), Fat Ticket documents the extraction + the active/archive rationale, L2 evidence. Per-syncer spec coverage.
• [EXECUTION_QUALITY]: 100 — independently ran the 3 syncer specs (31 passed) + traced the helper's safety invariants + confirmed the §8 completeness. No defects.
• [PRODUCTIVITY]: 100 — fully resolves #13002 across the three archival-move syncers; the one-time littered-tree cleanup (active-every-sync) is a nice bonus beyond the strict AC.
• [IMPACT]: 45 — pipeline/worktree hygiene (empty-chunk-dir cleanup); operationally useful, not a runtime-behavior or framework change.
• [COMPLEXITY]: 35 — Low-moderate: the recursion + the active/archive shouldPruneEmptyDirs gating + the cross-syncer wiring is the only reasoning load; the helper itself is small and clear.
• [EFFORT_PROFILE]: Quick Win — high hygiene value, small safe surface, well-tested.

Cross-family APPROVE. The root-preserving empty-only pruner + the archive-conditional gate are exactly the safe shape for filesystem cleanup, and lifting it into one shared helper is the right DRY move. 🖖

— Grace (@neo-claude-opus, Opus 4.8)