Skip to content

CLAUDE.md

Latest commit

 

History

History
88 lines (63 loc) · 3.83 KB

CLAUDE.md

File metadata and controls

88 lines (63 loc) · 3.83 KB
Raw

CLAUDE.md

@AGENTS.md

AGENTS.md above holds this repository's shared rules and is the source of truth. To change a rule, edit AGENTS.md. Keep this file limited to Claude Code specifics.

Language

Everything written into this repository is English — code, comments, docs, commit messages, issues, and PRs. See the language policy at the top of AGENTS.md.

The user may talk to you in Korean or any other language. Reply in whatever language they use, but what you write to disk, to git, and to GitHub is always English.

Skills

Project skills live in .agents/skills/, not .claude/skills/. .claude/skills is a symlink to .agents/skills; the real files are all under .agents/. To add a skill, create .agents/skills/<name>/SKILL.md.

Skill When to use
gitlab-api-endpoint Adding or changing a GitLab endpoint in packages/gitlab_api
feature-slice Adding a new screen or feature under apps/labfox/lib/features/
design-system-component Adding a shared widget to packages/design_system
responsive-screen Branching a screen across mobile / tablet / desktop

Reference docs

Read these only when relevant — do not load them into context by default.

  • .agents/docs/architecture.md — layer responsibilities, data flow, caching strategy
  • .agents/docs/conventions.md — naming, file placement, freezed/Riverpod rules
  • .agents/docs/api-reference.md — GitLab endpoint map, auth, pagination, terminology
  • .agents/docs/monetization.md — per-OS distribution, free/paid boundary, store entitlement
  • .agents/docs/billing.md — the subscription product, console values, purchase-to-entitlement flow
  • .agents/docs/roadmap.md — M0–M4, vertical slice, 1.0 scope
  • .agents/docs/workflow.md — Issue → Branch → PR procedure, gh commands, permission boundaries
  • .agents/docs/references.md — reference repos and docs, priority order, license warnings

Workflow

The origin remote is GitHub (theorvane/labfox). The CLI is gh.

Issue → Branch → Commit (-s) → Push → PR → Merge (maintainer)

⚠️ LabFox is a GitLab client but is developed on GitHub. App domain vocabulary is Merge Request; the development process is Pull Request. Do not mix them.

Fine without asking: creating issues, creating branches, committing, pushing, opening PRs. Never: pushing directly to main, force pushing, merging a PR, deleting issues/PRs/branches.

Subagents

Use of the Agent tool is permitted. You do not need to ask each time.

Good fits:

  • Surveying reference repositories (OctoLab, LabCoat, …) — read-only and high volume
  • Checking several GitLab API endpoints in parallel
  • Implementing independent feature slices
  • Broad code exploration (Explore)
  • PR review / code inspection

Rules:

  • Tell the subagent in its prompt to read and follow AGENTS.md — especially the language policy, the license rules (§8), the id/iid distinction, and the 1.0 scope (§9).
  • Never run parallel agents that touch the same files. Isolate with isolation: "worktree" or run them sequentially.
  • Do not delegate pushing or merging to a subagent.
  • Do not take a subagent's output at face value. Verify API paths and licenses yourself.
  • Launch independent work concurrently in a single message.

Habits

  • Work test-first (AGENTS.md §5a): write the failing test before the code, and for a bug reproduce it as a failing test first. A behaviour change without a test that would have failed before it is not done.
  • Read neighboring files at the same layer before writing code, and match their patterns.
  • Propose new dependencies and get approval first (AGENTS.md §4).
  • Touching a model is not done until build_runner has been run.
  • Report flutter analyze / flutter test output as it is. A failure is reported as a failure.