Scheduled MAX_OPEN_PRS gate in 5 workflow templates fails with exit code 4 when no matching PRs exist

[Evangelink]

Summary

Five of the upstream workflow templates have an unconditional pre_activation step that fails its scheduled runs with exit code 4 once gh pr list --search returns no matching PRs.

Affected templates

• workflows/adhoc-qa.md
• workflows/efficiency-improver.md
• workflows/glossary-maintainer.md
• workflows/perf-improver.md
• workflows/test-improver.md

All five use the same pattern (with different title prefixes):

MAX_OPEN_PRS=8
if [[ "$GITHUB_EVENT_NAME" != "schedule" ]]; then exit 0; fi
COUNT=$(gh pr list --repo "$GITHUB_REPOSITORY" --state open --search 'in:title "[<id>]"' --json number --jq 'length')
[[ "$COUNT" -lt "$MAX_OPEN_PRS" ]]

Root cause

gh pr list --search exits with code 4 when the search returns no matches (cli/cli "no results" exit code). The step runs under bash -e, so the command-substitution assignment to COUNT aborts the entire script with that same exit code, the pre_activation job is marked failed, and the workflow run is reported as a failure — even though "zero matching open PRs" is precisely the case where we want the agent to run.

This causes the workflows to fail every single scheduled run for repos that have no open PRs from these agents (which is most of the time after PRs get merged/closed).

Reproducer

In any repo where no PR title starts with [efficiency-improver]:

set -e
COUNT=$(gh pr list --state open --search 'in:title "[efficiency-improver]"' --json number --jq 'length')
echo "COUNT=$COUNT"   # never reached

The script aborts before printing, with exit code 4.

Examples in the wild

Daily failures in microsoft/testfx, all with Process completed with exit code 4 in the pre_activation Run MAX_OPEN_PRS=8 step:

• https://github.com/microsoft/testfx/actions/runs/26605466020/job/78399407141 (Efficiency Improver)
• https://github.com/microsoft/testfx/actions/runs/26620617306 (Glossary Maintainer)
• https://github.com/microsoft/testfx/actions/runs/26608040487 (Test Improver)

We're fixing it locally in microsoft/testfx#8676 but the bug originates here so adopters will keep hitting it.

Proposed fix

Fall back to COUNT=0 when gh pr list returns a non-zero exit code:

- COUNT=$(gh pr list --repo "$GITHUB_REPOSITORY" --state open --search 'in:title "[efficiency-improver]"' --json number --jq 'length')
+ # gh pr list exits with code 4 when --search returns no matches; fall back to 0 so set -e doesn't kill the script.
+ COUNT=$(gh pr list --repo "$GITHUB_REPOSITORY" --state open --search 'in:title "[efficiency-improver]"' --json number --jq 'length' 2>/dev/null || echo 0)
  [[ "$COUNT" -lt "$MAX_OPEN_PRS" ]]

This restores the intended semantics — the workflow runs when there are fewer than MAX_OPEN_PRS matching open PRs, and only short-circuits when the cap is actually reached.

Happy to send a PR if useful.

[Evangelink]

Update from the downstream PR (microsoft/testfx#8676): the original 2>/dev/null || echo 0 proposal got a fair review comment that it masks any gh failure (auth outage, API error, rate limit) and would silently let scheduled runs proceed. A safer version only swallows exit code 4 (no --search matches) and propagates anything else:

set +e
COUNT=$(gh pr list --repo "$GITHUB_REPOSITORY" --state open --search "$SEARCH" --json number --jq 'length' 2>/dev/null)
rc=$?
set -e
case $rc in
  0) ;;
  4) COUNT=0 ;;
  *) echo "gh pr list failed with exit code $rc" >&2; exit $rc ;;
esac
[[ "$COUNT" -lt "$MAX_OPEN_PRS" ]]

Happy to open a PR against the templates with this version if helpful.