Fix the auto-evaluation triggering in pr review agentic workflows

[JanKrivanek]

Fix: pr-triage worker no longer skips evaluate-now due to placeholder pr-status runs

Symptom. PRs were correctly classified as ready-for-eval and labeled pr-state/ready-for-eval, but the evaluate-now label was never applied, so evaluation.yml's label-driven entry point never fired. Repro on #703 (head a8bb6fc) and #720 (head 638726129): every hourly worker run logged eval-trigger: a workflow run already exists for <sha> — skipping.

Root cause. eval_run_exists_for_head() in .github/scripts/pr-triage-act.sh counted every evaluation.yml run for the head SHA. But evaluation.yml always posts a pr-status run (event=pull_request) on each push just to set the initial commit status — that run is not a real evaluation trigger. Its presence made the gate think evaluation had already run, so the worker skipped applying evaluate-now.

Fix. Filter the runs query to only events that actually trigger evaluation:

• issue_comment (/evaluate)
• workflow_dispatch (manual)
• pull_request_target with conclusion != "skipped" (the label-driven gate job runs; the status-only fork variant ends skipped)

Verification.

• Stuck SHAs (a8bb6fc, 638726129): filtered count = 0 → worker will now apply evaluate-now.
• SHA with a real /evaluate run (01050ea83): filtered count = 5 → gate still suppresses re-trigger as intended.

Unsticking existing PRs. This change ships in main, but worker checkouts already use the default branch, so the next hourly batch (cron :17) will fire evaluation on all currently-stuck pr-state/ready-for-eval PRs automatically. No manual label nudge required.

[Copilot]

Pull request overview

This PR fixes the PR-triage worker’s “evaluation already ran” gate so it doesn’t get tripped by placeholder evaluation.yml runs that only post an initial evaluation-status commit status, allowing the evaluate-now label to be applied and the label-driven evaluation entry point to fire.

Changes:

• Update .github/scripts/pr-triage-act.sh to count only evaluation workflow runs that represent real evaluation triggers (vs status-only placeholder runs).
• Switch the workflow-run query to a paginated scan with a jq filter over event types/conclusions.

Show a summary per file

File	Description
.github/scripts/pr-triage-act.sh	Refines the “evaluation run exists” detection logic used by the triage worker before applying evaluate-now.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 1/1 changed files
• Comments generated: 1

[github-actions]

👋 @JanKrivanek — this PR has 1 unresolved review thread(s). When you're ready, please address the feedback and push an update; the triage bot will pick up the next state automatically. (Add the no-stale label to silence further pings.)