docs(592): record that a skipped CI context is not red
#593
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: timothy/ersatztv#593
Reference in New Issue
Block a user
fixes #592
Docs-only. Amends
ci.monitor-armed-at-pr-openindocs/decisions/workflow-process.md(prose +Signals:only — heading andkey:unchanged, no supersession,[decisions-edit]in the commit).Why
Landing #436/#583 today, my CI monitor enumerated per-context statuses and treated anything
!= "success"as red, so it announced "NOT all green" on two fully green PRs — becauseBuild & push image (amd64)reportsskipped. Nothing was blocked (Gitea's combined/statustreatsskippedas non-blocking and reportedoverall=success, and both merged cleanly), but the monitor is what an agent reads to decide whether to investigate, so a false red costs a diagnosis cycle each time.Two corrections
if: github.event_name != 'pull_request'; images build only on push-to-main and tags. Not the docs-only mechanism. I initially misattributed it to #416's docs-only skip, which is a plausible-sounding wrong diagnosis: docs-only deliberately gates individual steps so required jobs still reportsuccessin seconds (ci.docs-only-skip-steps).decisions.mdalready records "buildisskippedon every PR" from the branch-protection angle — the missing piece was the monitor-authoring consequence.skipped/failure/cancelledare three meanings, not one. Settled-not-applicable; a real red; no verdict. Prefer gating on the combined.state, which already gets this right.The snippet was itself broken first — that's the point
My first draft documented
select(.status != "success" and .status != "skipped")after the jq pipeline had renamed.statusto.st. It compared againstnull, passed everything, and reported a fully green PR as nine failures. The committed version is verified in both directions:skippedbuild) ✅That two-way check is
process.bom-format-detection-recipe's "verify your detector" rule, which exists because a detector that can only say "ok" is worse than none.Review
Pure docs leaf, no server-state effect —
process.independent-review-rubricpermits a skip, but the record carries factual claims about CI wiring, so a bounded verification pass is running against the workflow file rather than taking my word for the job-levelif.Note
decisions_validate.pynow warnsaggregate active-corpus is 5606 lines (budget 5600) — schedule a consolidation. Non-blocking; flagged in #592 for its own pass.skippedCI context is not red [decisions-edit]Bounded independent fact-check (read-only, cheapest tier at bounded scope — proportionate to a pure-docs leaf per
process.independent-review-rubric, but run rather than skipped because the record makes factual claims about CI wiring). Not a style pass; accuracy only.All six claims verified TRUE, 4 files read:
Build & push image (amd64)is gated at the JOB level —.gitea/workflows/docker-build.yml:518,if: github.event_name != 'pull_request'on thebuildjob. Confirms it is skipped on every PR regardless of content.steps.detect.outputs.docs_only != 'true', lines 534/557/565/573); the job runs regardless — confirming the two mechanisms are genuinely different, which was the misdiagnosis this record exists to prevent.ci.docs-only-skip-stepsexists,status: active, and contains the exact phrase "buildisskippedon every PR" (docs/decisions.md:1699, key at :1677).ci.cancelled-is-not-a-verdict,ci.killed-job-triage,ci.infra-shaped-red-under-load,process.bom-format-detection-recipe. This was the defect I most wanted caught.key:/status:/since:/supersedes:/superseded-by:line is byte-identical toorigin/main, and the##heading is unchanged — so thedecisions lifecyclejob sees a prose edit, not an unlogged removal.[success, skipped, pending, failure]it emits onlypendingandfailure.No untrue claims, no syntax errors, no dangling references.
Review-verdict: MERGEABLE @
2fee3f0b94