Skip to content

CLAUDE.md

Latest commit

 

History

History
156 lines (108 loc) · 5.76 KB

CLAUDE.md

File metadata and controls

156 lines (108 loc) · 5.76 KB

CLAUDE.md — Monorepo egapro

Reference for all AI agents or developers working on this repository.

Business context

EGAPRO is the French government platform for declaring gender pay equity indicators (7 indicators, detailed in README.md).

Key concepts for development:

  • Indicators A–F: pre-calculated by GIP-MDS from DSN data, available each March
  • Indicator G: company-calculated pay gap by job categories (base + variable compensation)
  • Alert threshold: gap >= 5% triggers additional obligations (second declaration, CSE opinion, joint assessment)
  • CSE opinion: PDF upload, companies >= 100 employees only, up to 4/year
  • Company sizes: < 50 (voluntary), 50–99 (triennial), >= 100 (annual + CSE required)

All business rules are centralized in packages/app/src/modules/domain/ as pure TypeScript functions. See packages/app/CLAUDE.md for details.

Full specs: https://github.com/SocialGouv/egapro/wiki/Spec-V2

Monorepo structure

egapro/
  packages/
    app/        <- Next.js application (all active code)
    api/        <- Empty placeholder
  .github/
    workflows/  <- CI/CD GitHub Actions

Package manager: pnpm workspaces (pnpm@10).

Convention loading rule

If you work in packages/app/ or on a file that depends on it, you must load and strictly follow:

packages/app/CLAUDE.md

This file contains all package conventions: stack, module structure, React/TypeScript rules, DSFR, accessibility, tests, environment variables, scripts.

Absolute rule

Never create a git commit, unless the user explicitly requests it.

Git hygiene

  • No Co-Authored-By in commits or PR bodies.
  • No sensitive data committed: .env, credentials, secrets, API keys. Verify before every push.

Language policy

The site is in French, but all code must be in English:

  • All comments must be in English
  • All component names must be in English
  • All function and variable names must be in English
  • User-facing text (content, labels, buttons, links) remains in French

MCP Servers (.mcp.json)

Three MCP servers are configured and must be used in the relevant contexts:

Server When to use Key tools
next-devtools Debugging, runtime errors, route inspection, Next.js docs nextjs_index, nextjs_call, nextjs_docs, browser_eval
dsfr Before writing any DSFR HTML get_component_doc, search_components, get_color_tokens
figma When implementing from a Figma design get_design_context, get_screenshot

Next.js DevTools is particularly important: use nextjs_docs to look up Next.js APIs (never guess from memory), and use nextjs_call(get_errors) to check runtime/compilation errors after changes.

See packages/app/CLAUDE.md for detailed usage instructions per MCP server.

Useful root scripts

pnpm dev:app              # starts the app in dev mode (port 3000)
pnpm build                # builds all packages
pnpm lint:check           # checks lint (CI)
pnpm format:check         # checks format (CI)
pnpm typecheck            # checks TypeScript types
pnpm test                 # runs all unit tests
pnpm test:e2e             # runs Playwright E2E tests (requires dev server on port 3000)
pnpm test:lighthouse      # runs Lighthouse CI audit (requires dev server on port 3000)
pnpm db:migrate           # applies Drizzle migrations
pnpm db:studio            # opens Drizzle Studio

Note: pnpm test:e2e and pnpm test:lighthouse require the dev server running on port 3000 (pnpm dev:app). Lighthouse accessibility must score 100% — it is configured as an error threshold in .lighthouserc.json.

Automatic quality gates

Quality checks run automatically after every code change — no command needed. See .claude/rules/automation.md for full details.

Gate When How
Validation After every task validator agent (typecheck + test + lint + format)
Structure After every task structural-auditor agent (16 rules: forms, schemas, DRY, imports…)
RGAA After every task rgaa-auditor agent on modified .tsx files
Security After every task security-auditor agent on modified server files
Domain layer While writing Inline rules (also enforced by hooks + structural-auditor)
PR review When on a PR branch check-pr-reviews.sh hook at session start + /review skill

Agents and skills

Agents (.claude/agents/) — 4 agents, all run automatically after every task

Agent Role
validator Typecheck + test + lint + format (parallel)
structural-auditor 16-rule structural audit (code quality, forms, schemas, DRY, imports…)
rgaa-auditor 13-theme RGAA accessibility audit
security-auditor OWASP Top 10 + RGS security review

Skills (.claude/skills/) — /analyse + /implement + /ship + /review

Skill Purpose
/analyse [#N] Analyze issue, explore codebase, generate implementation plan.
/implement [#N] Fetch issue, create branch, code, run 4 validation agents.
/ship Create PR (single/split).
/review Address PR review comments (human + bots), fix, re-validate.

Workflow: /analyse #42 to plan, then /implement to code and validate, then /ship to create the PR, then /review to handle review comments.

CI

GitHub Actions workflows are in .github/workflows/ :

File Trigger Role
ci.yaml each push build · lint · format · typecheck · tests
release.yml manual (branch beta) semantic-release
review.yaml PR review app deployment
preproduction.yaml push beta preprod deployment
production.yaml push master prod deployment