feat(ci): add doc review pipeline with auto-labeling and Copilot review#6890
feat(ci): add doc review pipeline with auto-labeling and Copilot review#6890jstirnaman merged 38 commits intomasterfrom
Conversation
Detailed plan for two interconnected systems: 1. Label system overhaul (22 automation-driven labels replacing 30+ inconsistent ones) 2. Doc review workflow (Claude visual review + Copilot structural review with screenshots) This is a plan document only — no implementation changes. https://claude.ai/code/session_01D5rLaHdQv9iBL55UEsdQFt
|
@claude split |
- product:v3-monolith: Core, Enterprise (single-node / clusterable) - product:v3-distributed: Cloud Serverless, Cloud Dedicated, Clustered - Updated auto-label path mappings to match content directory structure - Updated migration mapping and label count (22 → 23) https://claude.ai/code/session_01D5rLaHdQv9iBL55UEsdQFt
jstirnaman
left a comment
There was a problem hiding this comment.
Keep agent primary instruction files consistent and lean.
- One CLAUDE.md (pointer) → role-specific files in .claude/agents/ - doc-triage-agent.md: label taxonomy, path mapping, priority rules - doc-review-agent.md: review scope, severity classification, output format - Prompt file (.github/prompts/) references agent file, stays workflow-specific - Updated file summary and implementation order https://claude.ai/code/session_01D5rLaHdQv9iBL55UEsdQFt
Claude now handles diff-only Markdown review (frontmatter, shortcodes, style, terminology). Copilot handles visual review by analyzing screenshots posted as images in PR comments. Key changes: - Job 3 (Claude) runs in parallel with Jobs 1→2→4 (diff-only, no screenshots) - Job 4 (Copilot) analyzes screenshots via @copilot PR comment mentions - Two prompt files: doc-review.md (Claude), copilot-visual-review.md (Copilot) - doc-review-agent.md scoped to diff-only (no screenshot analysis) - Q1 resolved: screenshots delivered to Copilot via PR comment images - Reduced Claude API cost (no image processing) - Added Copilot failure handling (fallback to human review of artifacts) https://claude.ai/code/session_01D5rLaHdQv9iBL55UEsdQFt
Convert Q2–Q5 from open recommendations to resolved decisions: - Q2: Advisory only (no required status checks) until false-positive rate confirmed - Q3: Playwright for CI screenshots, Puppeteer for local debugging - Q4: Poll preview URL with 15s interval and 10-min timeout - Q5: Cost acceptable with existing mitigations (path filters, skip-review, concurrency) Rename section from "Open Questions" to "Decisions (Resolved)". https://claude.ai/code/session_01D5rLaHdQv9iBL55UEsdQFt
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
|
@jstirnaman I've opened a new pull request, #6893, to work on those changes. Once the pull request is ready, I'll request review from you. |
…ements (#6893) * Initial plan * fix: resolve label naming inconsistency and document workflow updates - Rename review:approved to approval:codeowner to avoid confusion with review/* labels - Add note explaining the distinct prefix to prevent implementor confusion - Document required workflow updates for sync-plugin-docs label migration - Specify exact files and line numbers that need updating Co-authored-by: jstirnaman <212227+jstirnaman@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: jstirnaman <212227+jstirnaman@users.noreply.github.com>
Add Phase 2-3 pipeline components: doc-review workflow (3-job architecture), Claude/Copilot review prompts, URL resolver script, triage and review agents, and label guide. Deduplicate AGENTS.md (254→96 lines) by removing content available in referenced docs. Remove duplicate sections from copilot-instructions.md (264→221 lines). AGENTS.md now contains only high-signal guidelines loaded every session: commands, constraints, style rules, product paths, and reference pointers.
…abel groups for each product. Product group labels will be used to assign reviewers and help with content checks.
Add auto-label workflow that applies product and source labels to PRs based on changed file paths, using data/products.yml as the source of truth. Add workflow-utils.js shared helper for product path matching.
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
|
@sanderson TLDR; this adds the following labels, migrates issues and PRs from the old labels, and then deletes the old labels. |
- Add workflow_dispatch with pr_number input to both workflows for manual testing and on-demand re-runs - Migrate sync-plugin-docs label references to source:sync - Add area:agents, area:ci, area:links, release:*, good-first-issue, source:feedback, waiting:pr to labels.yml - Update products.yml: influxdb_cloud label_group v2 -> v2-cloud - Track label renames and deletions in LABEL_GUIDE.md
npm ci fails in sparse checkout because package-lock.json is not included. The workflow only needs js-yaml for YAML parsing.
Reverse the revert from 2f8efd6 to provide content changes that exercise the auto-label and doc-review workflows end-to-end.
|
|
@copilot Fix the broken links reported by link-checker: |
|
@jstirnaman I've opened a new pull request, #6910, to work on those changes. Once the pull request is ready, I'll request review from you. |
Visual Review SkippedThe PR preview deployment did not become available within 10 minutes. To trigger visual review manually, re-run this workflow after the |
curl --head outputs response headers before the status code from -w, so STATUS contained "HTTP/2 200 ...200" instead of just "200". Drop --head and add -o /dev/null to capture only the status code.
Preview Pages for Review2 page(s) changed in this PR: Preview URLs@copilot please review the preview pages listed above using the template below: Visual Review PromptReview the rendered documentation pages at the preview URLs listed below. ChecklistFor each preview URL, verify:
OutputFollow the shared review comment format, severity definitions, and label Adapt the "Files Reviewed" section to list preview URLs instead of file Preview URLs |
* Initial plan * fix: correct broken links in serve.md files for enterprise config-options Co-authored-by: jstirnaman <212227+jstirnaman@users.noreply.github.com> * Update content/influxdb3/enterprise/reference/cli/influxdb3/serve.md --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: jstirnaman <212227+jstirnaman@users.noreply.github.com> Co-authored-by: Jason Stirnaman <jstirnaman@influxdata.com>
Summary
product:*labels based on changed file paths (readsdata/products.yml)data/products.ymlwithcontent_pathandlabel_groupfields as single source of truth.github/instructions/content-review.instructions.md(auto-loaded by Copilot forcontent/**/*.md).github/templates/review-comment.md.github/copilot-instructions.mdby removing duplicated content.github/LABEL_GUIDE.mdand agent instructions in.claude/agents/product:*,source:*,waiting:*,area:*,review:*,release:*)Architecture
Both workflows support
workflow_dispatchwith apr_numberinput for on-demand re-runs.Label migration
Migrated labels to namespaced taxonomy:
AI assistant toolingarea:agentsci:testing-and-validationci:testingdesignarea:site-uiInfluxDB Cloudproduct:v2-clouduser feedbacksource:feedbackDeleted:
Pending PR(replaced bywaiting:pr),broke-link(replaced byarea:links)New labels:
waiting:pr,area:agents,area:ci,area:links,area:site-ui,product:v2-cloud,source:feedback,release:pending,release:ready,good-first-issueEnd-to-end test results
Triggered via
workflow_dispatchwithpr_number=6890:copilot-reviewsadded as reviewer)Visual review (Job 3) needs a content-changing PR to fully exercise preview URL polling.
Plan
See
.github/DOC-REVIEW-PIPELINE-PLAN.mdfor the full implementation plan.