You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Pulled the last ~50 ci.yml runs and categorized failures:
PR-branch failures (vast majority) β Lint Markdown (MD051 broken link
fragments) or Lint YAML (empty-lines trailing blank) errors in PR content.
The CI workflow is correctly catching real lint issues; not workflow bugs.
3 push-to-main failures (2026-06-08) β dependabot-rebase-reusable.yml:185
trailing blank line. Already fixed in main (file is now 184 lines, terminates
cleanly). No new remediation needed.
No infra-flake pattern observed β no checkout timeouts, action download
failures, or transient errors. Local re-run of every lint suite on main
passes cleanly (yamllint, markdownlint-cli2, actionlint, shellcheck).
Root Cause of Inflated Failure Rate
The Lint job runs three sub-steps sequentially (markdown β yaml β actions).
When the first step fails, the remaining two are short-circuited. PR authors
fix one issue, push, get the next failure, push, get the next β every
intermediate push counts as a fresh failed CI run in the Fleet Monitor window.
Scope
Apply if: always() to the Lint YAML and Lint GitHub Actions steps so all
three lint sub-checks run on every CI invocation. PR authors see every lint
issue in one report and can fix them in a single push, reducing the
fixβpushβfail loop that inflates the workflow's failure rate. The Lint job
name is preserved so the existing branch-protection required check is
unaffected.
Implementation Checklist
.github/workflows/ci.yml β add if: always() to the Lint YAML step
.github/workflows/ci.yml β add if: always() to the Lint GitHub Actions step
Tests to Write
TDD exception applies β this is a behavior-preserving CI-configuration
change. The gate semantics are identical (job still fails if any sub-step
fails); only the diagnostic completeness improves. Validation is by local
execution of every lint suite plus .dev-lead/scripts/dev-lead-lint.sh.
Test Command
yamllint, npx markdownlint-cli2 '**/*.md', actionlint, shellcheck, and bash .dev-lead/scripts/dev-lead-lint.sh
Triaged the last ~50 ci.yml runs and root-caused the failures
.github/workflows/ci.yml β added if: always() to Lint YAML
.github/workflows/ci.yml β added if: always() to Lint GitHub Actions
Verified locally with yamllint, markdownlint-cli2, actionlint, shellcheck, and .dev-lead/scripts/dev-lead-lint.sh
Triage Findings
Cohort
Count (sample of recent failed runs)
Root cause
PR-branch Lint Markdown failures
~20
Real MD051 broken-link-fragment issues in PR content
PR-branch Lint YAML failures
~10
Real empty-lines trailing-blank issues in PR content
main push failures (2026-06-08)
3
Same dependabot-rebase-reusable.yml:185 trailing blank β already fixed in main
Infra flakes
0
None observed
The workflow itself is correctly detecting real lint issues. The high failure
rate is inflated because, when one of the three Lint sub-steps fails, the
remaining two are skipped β so PR authors needed multiple fixβpush cycles to
discover every lint error, with each push counted as a fresh failed CI run.
.github/workflows/ci.yml β added if: always() to the second and third
steps of the Lint job (Lint YAML, Lint GitHub Actions) plus an
explanatory comment. The Lint job name is unchanged, so the existing Lint required status check in branch protection continues to work.
Notes
TDD exception applied. This is a behavior-preserving CI-configuration
change. The gate semantics are identical (the job still fails if any
sub-step fails); only the diagnostic completeness improves.
Edge case acknowledged. If actions/checkout itself fails, if: always()
would attempt to run the lint steps anyway, producing noisier output before
failing. This is an unusual failure mode and the noisier signal is still
obviously a checkout failure.
Follow-up to consider, not in scope here. The Fleet Monitor's
metric does not de-duplicate consecutive failed runs on the same PR branch;
iterative PR pushes inflate the rate. If that behavior is undesirable, the
fix belongs in the Fleet Monitor itself, not ci.yml.
Activity
don-petry commentedon Jun 9, 2026
Dev-Lead Implementation Plan
Issue: #329 β [Fleet Monitor] petry-projects/.github β ci.yml
Triage Summary
Pulled the last ~50
ci.ymlruns and categorized failures:Lint Markdown(MD051 broken linkfragments) or
Lint YAML(empty-linestrailing blank) errors in PR content.The CI workflow is correctly catching real lint issues; not workflow bugs.
dependabot-rebase-reusable.yml:185trailing blank line. Already fixed in main (file is now 184 lines, terminates
cleanly). No new remediation needed.
failures, or transient errors. Local re-run of every lint suite on
mainpasses cleanly (yamllint, markdownlint-cli2, actionlint, shellcheck).
Root Cause of Inflated Failure Rate
The
Lintjob runs three sub-steps sequentially (markdown β yaml β actions).When the first step fails, the remaining two are short-circuited. PR authors
fix one issue, push, get the next failure, push, get the next β every
intermediate push counts as a fresh failed CI run in the Fleet Monitor window.
Scope
Apply
if: always()to theLint YAMLandLint GitHub Actionssteps so allthree lint sub-checks run on every CI invocation. PR authors see every lint
issue in one report and can fix them in a single push, reducing the
fixβpushβfail loop that inflates the workflow's failure rate. The
Lintjobname is preserved so the existing branch-protection required check is
unaffected.
Implementation Checklist
.github/workflows/ci.ymlβ addif: always()to theLint YAMLstep.github/workflows/ci.ymlβ addif: always()to theLint GitHub ActionsstepTests to Write
change. The gate semantics are identical (job still fails if any sub-step
fails); only the diagnostic completeness improves. Validation is by local
execution of every lint suite plus
.dev-lead/scripts/dev-lead-lint.sh.Test Command
yamllint,npx markdownlint-cli2 '**/*.md',actionlint,shellcheck, andbash .dev-lead/scripts/dev-lead-lint.shdon-petry commentedon Jun 9, 2026
Dev-Lead: Implementation Complete
Plan Execution
ci.ymlruns and root-caused the failures.github/workflows/ci.ymlβ addedif: always()toLint YAML.github/workflows/ci.ymlβ addedif: always()toLint GitHub Actions.dev-lead/scripts/dev-lead-lint.shTriage Findings
Lint MarkdownfailuresLint YAMLfailuresempty-linestrailing-blank issues in PR contentmainpush failures (2026-06-08)dependabot-rebase-reusable.yml:185trailing blank β already fixed in mainThe workflow itself is correctly detecting real lint issues. The high failure
rate is inflated because, when one of the three Lint sub-steps fails, the
remaining two are skipped β so PR authors needed multiple fixβpush cycles to
discover every lint error, with each push counted as a fresh failed CI run.
Test Results
Files Changed
.github/workflows/ci.ymlβ addedif: always()to the second and thirdsteps of the
Lintjob (Lint YAML,Lint GitHub Actions) plus anexplanatory comment. The
Lintjob name is unchanged, so the existingLintrequired status check in branch protection continues to work.Notes
change. The gate semantics are identical (the job still fails if any
sub-step fails); only the diagnostic completeness improves.
actions/checkoutitself fails,if: always()would attempt to run the lint steps anyway, producing noisier output before
failing. This is an unusual failure mode and the noisier signal is still
obviously a checkout failure.
metric does not de-duplicate consecutive failed runs on the same PR branch;
iterative PR pushes inflate the rate. If that behavior is undesirable, the
fix belongs in the Fleet Monitor itself, not
ci.yml.don-petry commentedon Jun 9, 2026
feat: implement issue #329 β [Fleet Monitor] petry-projects/.github ββ¦
feat: implement issue #329 β [Fleet Monitor] petry-projects/.github ββ¦