Dev Skills

Fix local review (Codex, Zen, notes) w/ verify. Not gh-pr-review-fix or passive PR monitoring.

Convex-first specs—model, API, rollout, verify. New Convex plans. Not backend audit (convex-audit).

Audits a repo, route, screen, or 3D scene for motion quality, covering design-token consistency, frame-rate and draw-call risk, reduced-motion coverage, and accessibility. Use to audit, critique, or find motion gaps. Returns a prioritized punch list rather than implementing fixes.

Conduct WCAG 2.2 accessibility audits with automated testing, manual verification, and remediation guidance. Use when auditing websites for accessibility, fixing WCAG violations, or implementing accessible design patterns.

AGENTS.md maintenance rules for deciding when durable repo guidance should be updated after implementation.

Accessibility engineering for product interfaces, from focus states and keyboard support to ARIA, forms, and screen readers. Use when building or reviewing UI components, modals, menus, forms, custom widgets, or when the user says make this accessible or reports keyboard or screen-reader issues. Triggers on accessibility, a11y, WCAG, aria, focus ring, focus-visible, focus trap, keyboard navigation, tab order, tabindex, screen reader, sr-only, aria-live, alt text, hit area, touch target, prefers-reduced-motion, autoplay, toast duration, skip link, semantic HTML, aria-label, form errors, disabled buttons, not keyboard accessible.

TanStack Start/Router/Query integration: QueryClient context, SSR/prefetch, cache ownership, invalidation, and streaming.

Expert guidance for auditing, refactoring, and designing machine learning Jupyter notebooks with production-quality patterns. Use when: (1) Analyzing notebook structure and identifying anti-patterns, (2) Detecting data leakage and reproducibility issues, (3) Refactoring messy notebooks into modular pipelines, (4) Generating templates for ML workflows (EDA, classification, experiments), (5) Adding reproducibility instrumentation (seeding, logging, env capture), (6) Converting notebooks to Python scripts, (7) Generating experiment summary reports. Triggers on: ML notebook, Jupyter audit, notebook refactor, data leakage, experiment template, ipynb best practices, notebook to script, reproducibility.

Fix production Sentry issues from CLI evidence, using the issue, event, trace, replay, log and explore commands. Use for Sentry errors, performance regressions, noisy groups, source-map failures and crashes.

Generate repo-local dependency upgrade packs under `.agents/plans/upgrade/topic-slug/` with a playbook, trigger prompt, operator mode, and canonical manifest. Use when Codex needs to research an anchor package, expand related dependencies, detect target repo context, and create a reusable upgrade folder before any implementation work starts.

Review UI code against the Vercel Web Interface Guidelines: a broad, terse pre-commit sweep across interaction, forms, layout, content, performance, and accessibility. Use for a fast checklist pass over a diff. For depth in any single domain, and for a holistic cross-discipline review, use the better-* suite instead.

Architect-level Streamlit development for building, refactoring, debugging, testing, and deploying Streamlit apps (single-page or multipage) with correct rerun/state/caching/fragments, AppTest-based testing, custom components v2, safe theming/CSS, security-by-default, and Playwright MCP end-to-end automation.

Researches a Convex component against a private app's live backend graph, asks scored one-at-a-time design questions, and produces a reusable adoption or rejection package with `PLAN.md` and `CODEX_FULL_PROMPT.md`.