feat(agent): scheduled re-gate sweep (#777)

[JSONbored]

Closes #777

What & why

This completes Phase 0 (#768) — the agent layer's deny-by-default trust scaffolding — with the scheduling rail.

GitHub fires no per-PR event when a base branch advances (and none when a sibling duplicate merges, or the focus manifest / repo settings change). So an open PR's Gittensory Gate verdict can quietly go stale with nothing to refresh it. An hourly cron sweep recomputes those verdicts.

How

• index.ts — the existing 30-min cron enqueues an hourly agent-regate-sweep fan-out job (top-of-hour tick), right next to the other hourly refreshes.
• queue/processors.ts
  • fanOutAgentRegateSweepJobs — enqueues one per-repo sweep for every repo that opted the agent in (isAgentConfigured = at least one acting autonomy level). Mirrors the signal-snapshot fan-out so each repo's sweep is its own bounded, retryable queue message.
  • sweepRepoRegate — recomputes the deterministic gate verdict (buildPullRequestAdvisory → evaluateGateCheck, the exact webhook path) for that repo's stalest open PRs and records an agent.sweep.regate audit event.
• settings/agent-sweep.ts — pure selectRegateCandidates: drops drafts + PRs updated within the last hour (already gated by their webhook), takes the 25 stalest. Idempotent + rate-aware.
• settings/autonomy.ts — isAgentConfigured (any acting action class).

Safety (Phase-0 posture)

• Advisory only. Nothing is published to GitHub — no check, comment, or label — and no PR is mutated. The sweep only writes an internal audit record of the recomputed verdict. The action layer (feat(agent): maintainer write-actions layer (request-changes/approve/merge/close/label) #778) is what will later turn a flagged verdict into a real action.
• Rides the feat(agent): action audit log + kill-switch + dry-run/shadow mode #776 kill-switch. A global (AGENT_ACTIONS_PAUSED) or per-repo (agentPaused) pause records a skip and recomputes nothing.
• Deny-by-default. Repos that never set an acting autonomy level are skipped entirely.

Scope

No new repo setting, migration, or config-as-code surface — reuses the autonomy / agentPaused / agentDryRun config from #773/#774/#776.

Tests

• test/unit/agent-sweep.test.ts — candidate selection: freshness skip, stalest-first ordering + cap, missing-timestamp handling, draft/closed exclusion, deterministic tie-break, unparseable-now, defaults.
• test/unit/autonomy.test.ts — isAgentConfigured.
• test/unit/queue.test.ts — 5 processor cases: fan-out filtering (acting vs non-acting), advisory recompute with both a flagged (missing linked issue) and a clean verdict, kill-switch skip, defensive no-ops (missing repo arg / un-configured repo), and quiet-when-nothing-stale.

New code 100% covered (statements + branches); full suite green (2037 passed).

---

With this merged, Phase 0 is complete (#773/#774/#775/#776/#777). Next is Phase 1 #778 — the real write-actions layer — which mutates live PRs and so will come with an explicit design + sign-off before any code.

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#844 is no longer open.	No action.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[ghost]

gittensory · advisory review

Reviewed 9 changed file(s) — two independent AI reviewers.

Changed files (9)

• src/index.ts
• src/queue/processors.ts
• src/settings/agent-sweep.ts
• src/settings/autonomy.ts
• src/types.ts
• test/unit/agent-sweep.test.ts
• test/unit/autonomy.test.ts
• test/unit/index.test.ts
• test/unit/queue.test.ts

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

Reviewer A · gpt-oss-120b — recommends ✅ merge
The PR introduces a scheduled re‑gate sweep for stale PRs, adding a new job type, helper functions, and comprehensive tests. The implementation respects existing pause mechanisms, stays advisory, and does not expose any forbidden public content.

Suggestions

• Add a unit test covering the global pause (isGlobalAgentPause) to verify the sweep is denied when the whole system is paused.
• Update the project documentation or changelog to mention the new agent-regate-sweep scheduled job.

Worth double-checking

• Ensure the new job type is included in any job‑type validation or routing that may exist elsewhere in the codebase.
• Confirm that resolveAgentActionMode correctly handles all pause/dry‑run combos in production.

Reviewer B · nemotron-3-120b-a12b — recommends ✅ merge
This PR implements a scheduled re-gate sweep feature (#777) that periodically recomputes advisory gate verdicts for stale open PRs in agent-configured repositories. The changes are well-structured, add comprehensive tests, respect the public/private boundary (advisory-only audit events), and follow existing patterns (e.g., mirroring signal-snapshot fan-out). The code is deterministic, includes safety checks (pauses, autonomy configuration), and avoids public-facing actions.

No blocking issues spotted.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.60%. Comparing base (6930213) to head (f2d2c5c).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #844      +/-   ##
==========================================
+ Coverage   96.59%   96.60%   +0.01%     
==========================================
  Files         100      101       +1     
  Lines       14244    14295      +51     
  Branches     5192     5208      +16     
==========================================
+ Hits        13759    13810      +51     
  Misses        105      105              
  Partials      380      380              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.