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
The 2/9 failures in the monitored window both come from PR #8
(worktree-implement-sprint-1), where the app-build quality matrix job
legitimately failed on npm run typecheck / npm run lint β real
code-quality errors caught correctly by CI, not workflow flakiness. The
high p95 (~23 min) is the successful full-matrix build (npm ci + npm rebuild better-sqlite3 + typecheck/lint/test/coverage across
ubuntu/macos/windows), not a hung step.
ci.yml is already hardened against every known transient-failure class,
each locked by a regression guard in scripts/test-ci-workflow.sh
(Checks 0β8): concurrency SHA-scoping (#380), yq checksum extraction (#430),
curl --retry (#438), per-job timeout-minutes (#470), gitleaks --config
(#474). No workflow defect caused the monitored failures, so there is no
change that would reduce the failure rate without masking legitimate
code-quality failures.
Scope
Lock the one genuine, currently-unguarded durable invariant that keeps the
Fleet Monitor failure/cancelled metrics honest: any matrix strategy must set fail-fast: false, so every OS leg reports independently instead of a single
failing leg cancelling its siblings (cascade-cancellation inflates the
cancelled-run count and hides per-OS signal). This is the convention the quality matrix already relies on; the guard prevents its regression. The
check is a no-op when the workflow declares no matrix (as on main today),
mirroring Checks 5 and 8.
Implementation Checklist
scripts/test-ci-workflow.sh β add Check 9: any strategy.matrix job must set strategy.fail-fast: false; no-op when no matrix present. Update header doc block.
Tests to Write
scripts/test-ci-workflow.test.sh β fixture: matrix job with fail-fast: false is ACCEPTED
scripts/test-ci-workflow.test.sh β fixture: matrix job with fail-fast: true is REJECTED
scripts/test-ci-workflow.test.sh β fixture: matrix job omitting fail-fast is REJECTED
scripts/test-ci-workflow.test.sh β fixture: workflow with no matrix passes (Check 9 no-op)
Test Command
bash scripts/test-ci-workflow.sh && bash scripts/test-ci-workflow.test.sh
(full workflow-guard suite mirrors the workflow-tests CI job)
The 2/9 failures in the monitored window were legitimate code-quality gate
failures, not workflow flakiness: both come from PR #8
(worktree-implement-sprint-1), where the app-build quality matrix job
correctly failed on npm run typecheck / npm run lint (real ESLint/tsc
errors in WIP code). The high p95 (~23 min) is the successful full-matrix
build (npm ci + npm rebuild better-sqlite3 + typecheck/lint/test/coverage
across ubuntu/macos/windows), not a hung step. ci.yml is already hardened
against every known transient class (concurrency SHA-scoping #380, curl --retry#438, yq checksum extraction #430, per-job timeout-minutes#470,
gitleaks --config#474), each locked by a regression guard. No workflow
change can reduce those failures without masking legitimate CI signal.
The one genuine, currently-unguarded durable invariant that keeps the Fleet
Monitor failure + cancelled metrics honest is matrix result-fidelity: the quality matrix relies on strategy.fail-fast: false so every OS leg reports
independently. GitHub defaults an absent/true fail-fast to cancelling the
siblings of the first failing leg β those cancelled legs land as cancelled runs
and hide per-OS signal. I locked this as Check 9, following the
no-op-when-not-applicable pattern of Checks 5 and 8 (no-op on main, which has
no matrix; enforced once the build matrix merges to main). Verified the guard
accepts the existing worktree-implement-sprint-1quality matrix unchanged.
Plan Execution
Added Check 9 to scripts/test-ci-workflow.sh (matrix jobs must set fail-fast: false; no-op when no matrix); updated header doc block
TDD: confirmed new reject cases fail before Check 9 existed (red), pass after (green)
Test Results
=== test-ci-workflow.sh (guard on main ci.yml) ===
PASS: 'gitleaks detect' loads a committed --config (.gitleaks.toml) ...
PASS: no matrix strategy in .github/workflows/ci.yml β fail-fast check not applicable
All checks passed.
=== test-ci-workflow.test.sh (27 fixture cases) ===
ok - matrix job with strategy.fail-fast: false is accepted
ok - matrix job with strategy.fail-fast: true is rejected
ok - matrix job omitting strategy.fail-fast is rejected
ok - workflow without a matrix strategy passes (Check 9 skipped)
All tests passed.
=== guard against worktree-implement-sprint-1 ci.yml (real quality matrix) ===
PASS: matrix job 'quality' sets strategy.fail-fast: false
All checks passed.
=== full workflow-guard suite (mirrors CI workflow-tests job) β all 10 scripts PASS ===
=== shellcheck (changed scripts) β clean ; .dev-lead/scripts/dev-lead-lint.sh β all checks passed ===
Files Changed
scripts/test-ci-workflow.sh: added Check 9 (every strategy.matrix job must set strategy.fail-fast: false; no-op when no matrix) + header doc entry fix: configure CodeQL to scan Python onlyΒ #6.
No .github/workflows/ci.yml change was warranted: main's ci.yml declares no
matrix, and the monitored failures were legitimate code-quality failures, not
workflow defects. Per the Phase 2 exception, the change is a behavior-preserving
regression-guard addition, fully TDD'd via fixtures.
Boolean handling (string-compares yq's unquoted output) intentionally mirrors
the existing cancel-in-progress check (Check 4) for consistency.
Repo has no package.json/npm lint at root; .dev-lead/scripts/dev-lead-lint.sh
(shellcheck + CODEOWNERS validation) is the applicable lint gate and passes.
No commit/push/PR performed β left to the CI workflow per instructions.
Activity
don-petry commentedon Sep 6, 2026
Dev-Lead Implementation Plan
Issue: #486 β [Fleet Monitor] petry-projects/TalkTerm β .github/workflows/ci.yml
Investigation summary
The 2/9 failures in the monitored window both come from PR #8
(
worktree-implement-sprint-1), where the app-buildqualitymatrix joblegitimately failed on
npm run typecheck/npm run lintβ realcode-quality errors caught correctly by CI, not workflow flakiness. The
high p95 (~23 min) is the successful full-matrix build (
npm ci+npm rebuild better-sqlite3+ typecheck/lint/test/coverage acrossubuntu/macos/windows), not a hung step.
ci.ymlis already hardened against every known transient-failure class,each locked by a regression guard in
scripts/test-ci-workflow.sh(Checks 0β8): concurrency SHA-scoping (#380), yq checksum extraction (#430),
curl
--retry(#438), per-jobtimeout-minutes(#470), gitleaks--config(#474). No workflow defect caused the monitored failures, so there is no
change that would reduce the failure rate without masking legitimate
code-quality failures.
Scope
Lock the one genuine, currently-unguarded durable invariant that keeps the
Fleet Monitor failure/cancelled metrics honest: any matrix
strategymust setfail-fast: false, so every OS leg reports independently instead of a singlefailing leg cancelling its siblings (cascade-cancellation inflates the
cancelled-run count and hides per-OS signal). This is the convention the
qualitymatrix already relies on; the guard prevents its regression. Thecheck is a no-op when the workflow declares no matrix (as on
maintoday),mirroring Checks 5 and 8.
Implementation Checklist
scripts/test-ci-workflow.shβ add Check 9: anystrategy.matrixjob must setstrategy.fail-fast: false; no-op when no matrix present. Update header doc block.Tests to Write
scripts/test-ci-workflow.test.shβ fixture: matrix job withfail-fast: falseis ACCEPTEDscripts/test-ci-workflow.test.shβ fixture: matrix job withfail-fast: trueis REJECTEDscripts/test-ci-workflow.test.shβ fixture: matrix job omittingfail-fastis REJECTEDscripts/test-ci-workflow.test.shβ fixture: workflow with no matrix passes (Check 9 no-op)Test Command
bash scripts/test-ci-workflow.sh && bash scripts/test-ci-workflow.test.sh(full workflow-guard suite mirrors the
workflow-testsCI job)don-petry commentedon Sep 6, 2026
Dev-Lead: Implementation Complete
Investigation outcome
The 2/9 failures in the monitored window were legitimate code-quality gate
failures, not workflow flakiness: both come from PR #8
(
worktree-implement-sprint-1), where the app-buildqualitymatrix jobcorrectly failed on
npm run typecheck/npm run lint(real ESLint/tscerrors in WIP code). The high p95 (~23 min) is the successful full-matrix
build (
npm ci+npm rebuild better-sqlite3+ typecheck/lint/test/coverageacross ubuntu/macos/windows), not a hung step.
ci.ymlis already hardenedagainst every known transient class (concurrency SHA-scoping #380, curl
--retry#438, yq checksum extraction #430, per-jobtimeout-minutes#470,gitleaks
--config#474), each locked by a regression guard. No workflowchange can reduce those failures without masking legitimate CI signal.
The one genuine, currently-unguarded durable invariant that keeps the Fleet
Monitor failure + cancelled metrics honest is matrix result-fidelity: the
qualitymatrix relies onstrategy.fail-fast: falseso every OS leg reportsindependently. GitHub defaults an absent/true
fail-fastto cancelling thesiblings of the first failing leg β those cancelled legs land as cancelled runs
and hide per-OS signal. I locked this as Check 9, following the
no-op-when-not-applicable pattern of Checks 5 and 8 (no-op on
main, which hasno matrix; enforced once the build matrix merges to
main). Verified the guardaccepts the existing
worktree-implement-sprint-1qualitymatrix unchanged.Plan Execution
scripts/test-ci-workflow.sh(matrix jobs must setfail-fast: false; no-op when no matrix); updated header doc blockscripts/test-ci-workflow.test.sh(acceptfalse; rejecttrue; reject omission; no-matrix no-op)Test Results
Files Changed
scripts/test-ci-workflow.sh: added Check 9 (everystrategy.matrixjob must setstrategy.fail-fast: false; no-op when no matrix) + header doc entry fix: configure CodeQL to scan Python onlyΒ #6.scripts/test-ci-workflow.test.sh: added 3 matrix fixtures and 4 cases (24β27) covering accept/reject/omit/no-op.Notes
.github/workflows/ci.ymlchange was warranted:main's ci.yml declares nomatrix, and the monitored failures were legitimate code-quality failures, not
workflow defects. Per the Phase 2 exception, the change is a behavior-preserving
regression-guard addition, fully TDD'd via fixtures.
the existing
cancel-in-progresscheck (Check 4) for consistency.package.json/npm lint at root;.dev-lead/scripts/dev-lead-lint.sh(shellcheck + CODEOWNERS validation) is the applicable lint gate and passes.
feat: implement issue #486 β [Fleet Monitor] petry-projects/TalkTerm β¦