[Fleet Monitor] petry-projects/google-app-scripts — .github/workflows/nodejs-tests.yml #388

Activity

1. don-petry

   added

   health-check

   dev-leadFor dev-lead agent pickupFor dev-lead agent pickup

   fleet-tracker

   on Jun 15, 2026
2. don-petry commented on Jun 15, 2026

   

   don-petry

   on Jun 15, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead Implementation Plan

   Issue: #388 — [Fleet Monitor] petry-projects/google-app-scripts — .github/workflows/nodejs-tests.yml

   Root-cause analysis

   • The 12 runs the monitor counts as failures all have conclusion action_required with zero jobs executed — they are PR runs gated behind GitHub's "approve workflow run" setting, not test failures. There are 0 failure conclusions in the window (68 success, 12 action_required). Approval-gating is a repository Actions setting and cannot be changed from a workflow file.
   • However, nodejs-tests.yml has a genuine reliability/correctness defect: the "Discover packages" step only finds top-level directories containing a tests/ subdir (deploy, gas-installer) and then runs npx jest "<pkg>/tests" with a loose path-regex. As a result CI executes only 87 of 301 tests — the calendar-to-briefing-doc, calendar-to-sheets, and gmail-to-drive-by-labels packages are never run in CI. The discovery also depends on fs.readdirSync ordering, making it non-deterministic.
   • coverage.yml already runs the full suite deterministically with npm test -- --coverage, confirming the brittle discovery here is unnecessary.

   Scope

   Make nodejs-tests.yml run the full Jest suite deterministically (matching the project's npm test entry point) and drop the unused elevated actions: write permission.

   Implementation Checklist

   • Replace the brittle "Discover packages" + per-package loop in .github/workflows/nodejs-tests.yml with a single deterministic npm test invocation
   • Remove the unused actions: write job permission (least privilege; matches coverage.yml)

   Tests to Write

   • tests/workflows/nodejs-tests-workflow.test.js — assert the workflow runs the full suite via npm test, does not use the brittle top-level discovery pattern, and does not grant actions: write

   Test Command

   npm test
3. don-petry commented on Jun 15, 2026

   

   don-petry

   on Jun 15, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead: Implementation Complete

   Root cause

   The 12 runs the Fleet Monitor counts as failures all have conclusion action_required with zero jobs executed — they are PR runs gated behind GitHub's "approve workflow run" setting (a repository Actions setting), not test failures. The window has 0 failure conclusions (68 success, 12 action_required). That gating cannot be changed from a workflow file.

   Separately, nodejs-tests.yml had a real reliability/correctness defect: the "Discover packages" step only found top-level directories containing a tests/ subdir (deploy, gas-installer) and ran npx jest "<pkg>/tests" with a loose path-regex. Net effect: CI executed only 87 of 301 tests — calendar-to-briefing-doc, calendar-to-sheets, and gmail-to-drive-by-labels were never run in CI — and discovery depended on non-deterministic fs.readdirSync ordering. coverage.yml already runs the full suite deterministically via npm test, confirming the brittle discovery was unnecessary.

   Plan Execution

   • Replaced the brittle "Discover packages" + per-package loop with a single deterministic npm test -- --colors (runs all packages via Jest testMatch)
   • Removed the unused actions: write job permission (least privilege; matches coverage.yml)
   • Added a regression test guarding the workflow against the brittle pattern

   Test Results

   > jest --runInBand
   
   
   Test Suites: 13 passed, 13 total
   Tests:       304 passed, 304 total
   Snapshots:   0 total
   Time:        1.691 s, estimated 4 s
   Ran all test suites.
   

   • Coverage: statements 98.91%, branches 90.00%, functions 98.00%, lines 100% — all thresholds met (node scripts/check-coverage.js).
   • npm run check (prettier + eslint): 0 errors.

   Files Changed

   • .github/workflows/nodejs-tests.yml: replaced FS-order-dependent discovery + loop (which silently skipped 3 of 5 test packages) with deterministic npm test -- --colors; dropped unused actions: write permission.
   • tests/workflows/nodejs-tests-workflow.test.js: new test asserting the workflow runs the full suite via npm test, avoids the brittle top-level discovery, and grants no actions: write.

   Notes

   • The monitor's "failure" signal is GitHub's action_required approval gate on PR runs — a repository setting outside any workflow file's control. The change here fixes the genuine workflow defects (non-deterministic, partial-coverage test execution) that left CI unable to actually validate most of the codebase.
   • .dev-lead/scripts/dev-lead-lint.sh reports pre-existing SC2043 shellcheck warnings in scripts/apply-repo-settings.sh and scripts/setup-code-quality-ruleset.sh. These files are unrelated to this issue and unchanged vs main; left untouched per the "do not modify unrelated files" constraint.
4. don-petry commented on Jun 15, 2026

   

   don-petry

   on Jun 15, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead: Lint Check Failed

   The implementation for issue #388 contained lint errors. The commit was aborted to prevent a CI failure.

     [lint] shellcheck on 4 file(s)...
   
   In scripts/apply-repo-settings.sh line 40:
   for cmd in gh; do
              ^-- SC2043 (warning): This loop will only ever run once. Bad quoting or missing glob/expansion?
   
   
   In scripts/setup-code-quality-ruleset.sh line 39:
   for cmd in gh; do
              ^-- SC2043 (warning): This loop will only ever run once. Bad quoting or missing glob/expansion?
   
   For more information:
     https://www.shellcheck.net/wiki/SC2043 -- This loop will only ever run once...
   

   To retry: fix the lint errors locally (or re-apply the dev-lead label — the agent will try again).
5. don-petry commented on Jul 15, 2026

   

   don-petry

   on Jul 15, 2026

   CollaboratorAuthor

   More actions

   ✅ Auto-resolved by Fleet Monitor. This workflow's failure rate has been below the 10% threshold (or the workflow was removed) since 2026-06-15 (30 days) — the monitor stopped flagging it. Closing. It will reopen automatically if the workflow crosses the threshold again.
6. don-petry

   closed this as completedon Jul 15, 2026