Repository navigation
docs: add an advisory PR-time doc-integrity check to docs-ui-tests - #2991
marcel-rbro wants to merge 4 commits into
Conversation
Splits the workflow into two jobs so doc-side drift is caught in review instead of up to a week later. doc-integrity runs on pull requests touching sources/platform/** or docs-tests/**. It runs only the `integrity` project — static checks that each assertion's source_file and source_quote still resolve. No browser and no credentials, so it needs no Playwright install and works on fork PRs, where secrets are unavailable. The check is informative only: `continue-on-error` keeps a failure from blocking the merge, and findings are written to the job summary as a table of assertion IDs and their doc sources, so a reviewer sees them without opening logs. Promoting it to a gate later means removing `continue-on-error` and marking it required in branch protection. ui-drift is unchanged except for an `if` guard keeping it off pull requests. It still runs weekly and on demand, against staging, and still files a drift issue on failure. The motivation is #2951: #2919 moved account/settings.md on 09-01, the baseline went stale, and nothing surfaced it until the scheduled run on 09-07 — into an unassigned issue that sat another 10 days. Run against that state locally, this job reports all 16 stale back-references in 4.3 seconds with no browser and no credentials. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FSj1mTUbMKgmu2JoGJLSKD
|
✅ Preview for this PR (commit |
|
I spy with my little 👀 a red PR check? |
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
TC-MO
left a comment
There was a problem hiding this comment.
Thanks, running the doc-side check on PRs fixes the gap that let #2951 go unnoticed for a week. I ran pnpm test:integrity locally with no browsers installed, and it works as described.
I'd like two changes before merge. Both matter because master is already stale: #2980 broke own-oauth-client-option, and #3016 tracks it. With the current wording, the summary on every docs PR would blame that PR for this failure. CI retries also list each finding twice. The other comments are optional suggestions.
- Compare against the base commit and split findings into new vs. already unresolved on the base branch, with neutral wording - Run with --retries=0 so CI's retry doesn't list findings twice - Emit a ::warning annotation per new finding - Summarize only on failure; drop the success branch - Run on changes to this workflow; cancel superseded PR runs - Shorten the header and job comments Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
|
We can wait with this for now. It's blocked by the captcha issue. |
Part of #2671. Addresses the cause behind #2988.
The harness only ran weekly, so #2919 moved a doc on 09-01 and nothing surfaced it until 09-07 — into an unassigned issue that then sat 10 days.
Splits the workflow into two jobs:
doc-integrity(new)sources/platform/**ordocs-tests/**source_file/source_quotestill resolvesui-drift(unchanged)Informative only:
continue-on-error: true, so a failure never blocks the merge. Findings render as a table in the job summary. Promoting it to a gate later means dropping that flag and marking the check required.No
playwright installand no secrets — the doc-side project never opens a browser, so it also works on fork PRs. Verified against the pre-#2988 state with no browser present: all 16 stale references reported in 4.3s.Note: this PR only touches
.github/workflows/, so the job won't self-trigger here. It first runs on the next docs PR after merge.🤖 Generated with Claude Code
https://claude.ai/code/session_01FSj1mTUbMKgmu2JoGJLSKD