Maintain Monorepo

Maintain: Monorepo + devantler.tech site

Repo devantler-tech/monorepo · path repo root + docs/ · Issues ENABLED (issue + PR triage/creation, labels, comments). Shared rules: monorepo AGENTS.md. Memory = your native memory (the monorepo/site cursors + caches, e.g. caches.md) + the end-of-run report; no version-controlled dashboard, no bespoke state.json.

Roadmap & enhancement: the site/repo roadmap lives in GitHub Issues here (roadmap-labelled epics + milestones). Beyond the maintenance menu below, advance the site via product-engineering: docs/site features, accessibility, performance (bundle size / Lighthouse), and content quality (the Site QA + Content Review tasks below are the recurring slice of that). The blog is a maintained public product: use privacy-safe evidence to improve discovery, comprehension, adoption, and portfolio positioning through both new posts and material refreshes. Validate with the docs build before any PR.

Repo-specific conventions

• Branch claude/<area>-<desc>-<issue> (the issue number makes a pre-PR claim matchable — see the contract's Claim protocol; issue-less hotfixes and trivial obvious fixes keep plain claude/<area>-<desc>). When grepping open PR bodies for #<issue>, keep only same-repo references (drop a hit whose only #<issue> is a foreign owner/repo#<issue> — a body naming both repos is still a valid local claim). Submodule pointers often show as modified — expected, not yours; detect real dirtiness with git status --porcelain --ignore-submodules=all; never stage pointer bumps; never git submodule update --remote.
• Validate: before any docs/ PR, [ -d docs/node_modules ] || (cd docs && npm ci) then cd docs && npm run build (use npm --prefix docs …). Workflows → actionlint. Never edit auto-generated files / *.lock.yml.
• Agent/review files: AGENTS.md is the single canonical instruction file — what humans, agents, and Copilot code review all read (since 2026-06-18); there is no separate .github/copilot-instructions.md (retired portfolio-wide). Keep AGENTS.md, the .claude/ skills/cards, and any path-scoped .github/instructions/ files from drifting apart (see product-engineering §7).
• Labels (apply only from this set): automation, documentation, github_actions, dependencies, submodules, bug, enhancement, question, duplicate, wontfix, needs triage, needs investigation, performance, refactor, security, repo-assist, roadmap, good first issue, help wanted, spam, blocked, next. Live taxonomy is the verification source (gh label list -R devantler-tech/monorepo); the allowlist above must stay a subset of it — never invent label names. Contract self-test: .claude/scripts/product-value-contract.test.sh parses every backticked allowlist label against a pinned live-taxonomy set (and still rejects the retired ci / agentic-workflows entries).

Task menu (1–3 highest-value; Content Review gated to Mondays; Blog Stewardship low priority)

• A. CI Doctor: gh run list --repo devantler-tech/monorepo --status failure --limit 20 ... (active workflows CI + Publish - Pages; ignore removed gh-aw workflows & Dependabot's submodule-update job). Dedupe vs native memory's CI-investigation cache (caches.md). gh run view <id> --log-failed (untrusted); root-cause (bad MDX/frontmatter, broken imports, npm ci lockfile desync, Node/Sharp, dead submodule ref, broken YAML). Failure on an open PR → one root-cause comment (+ draft fix if confident); else record it in native memory (caches.md). Prune cache >7d.
• B. CI/CD health: workflows ci.yaml, publish-pages.yaml, sync-labels.yaml, todos.yaml, copilot-setup-steps.yml. Never run gh aw / recompile *.lock.yml. Let exact Renovate/Dependabot PRs self-progress only while the root contract's live evidence qualifies; when one cannot finish autonomously, diagnose, repair, and merge it as trusted rung-one work (majors included). Safe improvements to dependency automation or workflows (caching, path filters, concurrency, pinned actions) → one issue-driven ci: draft PR; validate YAML first.
• C. Site QA — rotate one sub-task/run (site_qa_cursor: link-check → accessibility → multi-device → …). Use npm --prefix docs; for browser checks start preview backgrounded, CAPTURE the PID, and ALWAYS kill it (free port 4321) even on failure. Link-check: extract URLs from docs/src/content, curl -sL -o /dev/null -w "%{http_code}"; skip the unfixable_links cache; broken → fix (draft docs: fix broken links) or cache with reason. Accessibility: heading hierarchy, alt text, descriptive link text, ARIA, skip-link, WCAG-AA green (#39ff14 dark / #15803d light). Multi-device (Playwright installed globally; chromium cached; if unavailable skip + advance cursor): iPhone 12/Pixel 5/iPad/iPad Pro/Desktop × /,/about/,/projects/,/templates/,/blog/ (note: / is a Starlight splash with no sidebar by design; a refused GA gtag request is the sandbox, not a bug).
• D. Content Sync (don't git submodule update --remote): per-project descriptions live in docs/src/content/docs/projects/active.mdx under ## headings (NOT projects/index.mdx). Map ksail→## KSail (brief + link to ksail.devantler.tech only — never duplicate KSail docs), platform→## Platform, actions→## Actions, reusable-workflows→## Reusable Workflows, go/dotnet-template→docs/src/content/docs/templates/{go,dotnet}.md. Private applications/* are intentionally NOT mapped. Never modify blog posts during Content Sync — blog changes use task G with their own evidence/editorial review. Build-verify → draft docs: sync project descriptions.
• E. Content Review (Mondays only; rotate unbloat ↔ editorial in content_review): Unbloat one not-recently-cleaned file (skip frontmatter disable-agentic-editing: true), cut duplication/verbosity ≥20% w/o losing accuracy. Editorial rotate lens weekly (Technical Rigor → Clarity → Onboarding → Portfolio Balance).
• F. Repo Assist: triage/label new issues + PRs; comment on the oldest open item lacking an AI comment (1–3/run); self-spotted fixes → draft PR (use Fixes #N when it closes an issue); ≤3 stale-PR/issue nudges.
• G. Blog Stewardship (low priority; monthly evidence review, worthwhile publish/refresh every 4–8 weeks): complete a topic/evidence audit across portfolio balance, recurring questions/issues, shipped outcomes, and stale claims before advancing last_blog_review. Inspect privacy-safe aggregate Umami, referral, and CTA signals when access exists; advance last_metrics_review only after an actual aggregate inspection. If access or instrumentation is unavailable, file or update one deduplicated issue, record needs_attention, and Do not advance last_metrics_review. Before opening a blog experiment, check for any open blog experiment issue and any open publish/refresh PR. Maintain that existing lane (CI, review, deployment, or due measurement) and start no new post while either is open. Derive publication due-ness from the later of last_blog_publish and last_blog_refresh; a review alone cannot postpone it. Every substantive publication or refresh uses an issue as its experiment record (audience, evidence/baseline/proxy, hypothesis, intended signal, measurement window, follow-up date, and resulting decision), plus a delivery child closed by the publication PR. Update last_blog_publish only after a new-post PR has merged and deployed, and last_blog_refresh only after a material-refresh PR has merged and deployed; last_blog_stewardship is the later of actual publish/refresh dates and never advances for a draft, open PR, failed deployment, or review-only work. Choose a worthwhile evidence-backed story, never filler. For shipped work use problem/context → why it matters → what Devantler Tech built or learned → verified outcome/trade-offs → clear next step. For a current initiative use problem → why now → current status, clearly separate shipped from planned work, state unknowns/trade-offs, and give the next step without claiming an outcome. Define jargon and portfolio context, avoid internal run diaries and fabricated experience/metrics, and balance products over time. Verify frontmatter, title/description/excerpt, tags, cover/alt, commands, versions, licenses, claims, examples, links, RSS inclusion, social/OG presentation, a measurable CTA, the production build, and a multi-device preview.