Handle keepalive pause and avoid auto relabel - #25
Conversation
Automated Status SummaryHead SHA: d52ff93
Coverage Overview
Coverage Trend
Updated automatically; will refresh on subsequent CI/Docker completions. Keepalive checklistScopeNo scope information available Tasks
Acceptance criteria
|
There was a problem hiding this comment.
Pull request overview
This PR adds support for pausing keepalive operations via an agents:pause label and removes automatic remediation behaviors (auto-labeling and auto-assignment) to give users more explicit control.
Key changes:
- Introduces
agents:pauselabel handling to halt keepalive operations - Removes auto-labeling logic that previously added missing keepalive labels automatically
- Removes auto-assignment logic that attempted to assign human owners to PRs lacking assignees
- Adds pre-gate check result handling to the orchestrator for better visibility
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
.github/scripts/keepalive_gate.js |
Adds PAUSE_LABEL constant and integrates pause label checking into the gate evaluation logic, returning hasPauseLabel status and setting ok=false with reason 'keepalive-paused' when detected |
.github/scripts/keepalive_orchestrator_gate_runner.js |
Adds pre-gate result handling, implements pause label detection, removes auto-labeling for missing keepalive labels, removes auto-assignment for missing human assignees, and changes no-assignee behavior from attempting assignment to skipping keepalive |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| if (agentAlias) { | ||
| requiredLabels.push(`agent:${agentAlias}`); | ||
| } | ||
| const missingLabels = requiredLabels.filter((label) => !currentLabels.has(label)); |
There was a problem hiding this comment.
The variable missingLabels is computed but never used. Since the auto-labeling logic has been removed, this line should be deleted as well to avoid unnecessary computation and improve code clarity.
| const missingLabels = requiredLabels.filter((label) => !currentLabels.has(label)); |
| if (currentLabels.has('agents:pause')) { | ||
| addReason('keepalive-paused'); | ||
| summary.addRaw('Keepalive paused by agents:pause label.').addEOL(); | ||
| } | ||
|
|
There was a problem hiding this comment.
The pause label check here is redundant with the pre-gate evaluation. The evaluateKeepaliveGate function already checks for the pause label and returns hasPauseLabel in its result (and sets ok=false with reason='keepalive-paused' when the label is present). This reason has already been added at line 167 via addReason(preGate.reason || 'pre-gate-failed'). Consider removing this duplicate check and instead use preGate.hasPauseLabel if needed, or rely on the pre-gate check result that's already being processed at lines 166-175.
| if (currentLabels.has('agents:pause')) { | |
| addReason('keepalive-paused'); | |
| summary.addRaw('Keepalive paused by agents:pause label.').addEOL(); | |
| } |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const MAX_RUNS_PREFIX = 'agents:max-runs:'; | ||
| const SYNC_REQUIRED_LABEL = 'agents:sync-required'; | ||
| const ACTIVATED_LABEL = 'agents:activated'; | ||
| const PAUSE_LABEL = 'agents:pause'; |
There was a problem hiding this comment.
The new pause handling checks for agents:pause, but the rest of the keepalive stack uses the agents:paused label (see scripts/keepalive-runner.js line 483 and the paused fixture in tests/workflows/fixtures/keepalive/paused.json). With the name mismatch, PRs already marked agents:paused will not be treated as paused by this gate, so keepalive runs continue even though the pause label is present.
Useful? React with 👍 / 👎.
… systems The opener cron selects work by `priority:high|normal|low` label. The weekly repo-review's design-vs-impl pass selects by traced gap. Issues created manually with `enhancement` or `feature` labels but no `priority:*` label fall into a hole: neither system sees them. Worked example surfaced 2026-05-07: Inv-Man-Intake #25, #26, #27 were created 2026-03-01 with `enhancement` + `milestone:B-extraction-queue-images` only. 71 days later, no activity, no agent ever picked them up. New `scripts/repo_review_backlog_scan.py`: - Scans all active repos via `gh issue list` - Surfaces issues with `enhancement` OR `feature` label that: - have NO `priority:*` label (opener would already see them) - have NO `agent:*` label (an agent owns it) - have NO dependabot / sync* / process-eval label (other automations) - have NOT been updated in the past `--stale-days` days (default 7) - Sorts oldest-stale-first; writes JSON to --out Coordinator runs the scan after the final evaluator pass and feeds the result to the notify step. The desktop file now includes a "Backlog needing your attention" section listing each stale item with the exact three `gh issue edit/close` commands to promote / deprioritize / close it. macOS notification subtitle shows both queue and backlog counts. Goal: an enhancement issue sits unaddressed for at most 1-2 weekly cycles before the human surfaces it. Smoke-test against current state: found 4 items in Inv-Man-Intake (#27, #26, #25 at 71 days; #10 at 14 days) — exactly the pattern the user's report identified. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…mans The 2026-05-10 feedback: surfacing every unaddressed enhancement issue for human triage is heavier than necessary. The system should label the clear-cut cases itself and only ask for human input when the issue shape is ambiguous. Two-tier classification in `repo_review_backlog_scan.py`: **auto-labelable** — leaf enhancement issues. Heuristic priority: - has `milestone:*` label → `priority:normal` (declared planned work) - created >90 days ago without milestone → `priority:low` (likely stale) - otherwise → `priority:normal` (default) With `--apply` (the cron default), the scanner calls `gh issue edit --add-label priority:X` for each leaf. The opener picks them up on its next pass. **needs-human** — surfaced for decision, not auto-labeled: - title contains epic/tracker/umbrella/roadmap/rollup/parent issue - has `epic`/`umbrella`/`tracker`/`meta`/`parent`/`needs-triage`/`discussion` - has any `blocked*` label - body has ≥2 task-checkbox lines mentioning #NNN (umbrella tracking children) - body has a "Children: #..." or "Child issues: #..." declaration Validated against the user's worked example: #25/#26/#27 → leaf (auto-label `priority:normal`); #10 → umbrella detected via 4 child checkboxes (surface for human); #7 → umbrella detected via "epic" in title. Matches the user's stated preference exactly. The desktop reminder now renders two backlog subsections: - **Auto-labeled this week (N items) — FYI, no action required**: brief one-line summary per item with applied priority. - **Backlog needing your decision (M items)**: full three-command resolution block per item (promote / deprioritize / close). macOS notification text now distinguishes the cases: clean weeks with auto-labels only show "no action required"; weeks with needs-human items show "decisions needed". Coordinator now passes `--apply` to the scanner so the cron actually mutates labels (dry-run is still the default when running the scanner standalone for testing). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
No description provided.