Skip to content

AGENTS.md

Latest commit

 

History

History
64 lines (37 loc) · 4.26 KB

AGENTS.md

File metadata and controls

64 lines (37 loc) · 4.26 KB

Attitude

Follow the literal request; it is the contract, not a hint at what the user "really" wants.

  • Scope is the user's to widen. Reinterpret the goal, weigh alternatives, or expand the task only on an explicit hand-off ("figure it out", "you decide"). Take a confident, specific ask as given.
  • Fidelity binds the goal, not the effort. Within that goal, act with full initiative: do the substeps it needs, verify your work, surface what you notice. Literal scope is no excuse for passive execution.
  • Default over ask. On an ambiguous detail, pick the sensible default and say what you chose; reserve questions for forks only the user can settle.

Skills

All conventions and workflows live as skills under .codex/skills/. Read the linked file when its topic applies.

Project Outline

What typia is, the package family, the JS-descriptor / Go-plugin boundary, the workspace layout, and the canonical commands, .codex/skills/project/SKILL.md.

Development

Work rules, testing, validation, change integrity, .codex/skills/development/SKILL.md. Read before writing or modifying code.

Documentation

READMEs and the website guides, .codex/skills/documentation/SKILL.md. Read before writing or modifying docs.

Multi-Agent Workflows

Review Cycle, Discussion, Research Review Round, .codex/skills/multi-agent/SKILL.md. Read the Briefing subagents rule before delegating to any subagent; read a mode in full only when the user asks for it by name.

Pull Request Submission

PR submission flow and the release reference, .codex/skills/pull-request/SKILL.md. Read only when the user explicitly asks for a pull request; never open, push, or propose a PR on your own initiative.

Issue PR Workflow

GitHub issue triage, .wiki handoff updates, branch-per-issue implementation, mandatory Research Review Round, CI gating, PR comments, and issue closure, .codex/skills/issue-pr-workflow/SKILL.md. Read before sweeping issues or running the issue-to-PR loop.

Benchmark

The @typia/benchmark runner, fixtures, and per-CPU result archive, .codex/skills/benchmark/SKILL.md. Read before running, modifying, or publishing benchmark results.

Maintenance

Writing style

AGENTS.md and SKILL.md files are read by humans as well as agents.

  • Concise means no redundancy, no padding: not the same as cramming long sentences into one dense paragraph.
  • Concise does not mean gutted. Drop repetition; keep the rule and the rationale that makes it usable.
  • Match structure to content. Bullets for parallel items, a short paragraph for a single idea, a code block for a command.
  • State the rule first, then the reason. Use negative phrasing only for named failure modes the affirmative does not already cover.
  • Skills point, not paraphrase. Don't restate what the website, READMEs, or source comments already say — link to them. Skills are for cross-cutting rules and conventions, not a second copy of project docs.

AGENTS.md

The single shared entry point for both Claude Code (via CLAUDE.md → @AGENTS.md) and Codex CLI; a table of contents, not content. The H2s are ## Attitude, ## Skills, and ## Maintenance. ## Attitude is the one place global agent-behavior rules live; everything else points to a skill.

Update only for repository-contract changes: a new skill area, a renamed or merged skill, a workflow that no longer fits an existing skill, a release-process change, or a coding-agent rule that applies globally before any skill loads.

Skills

  • Location. .codex/skills/<kebab-name>/SKILL.md. No numeric prefix. Each file starts with YAML frontmatter holding name (the kebab folder name) and description (the one-line pointer Codex shows); Claude Code only auto-discovers .claude/skills/, so AGENTS.md still points to each SKILL.md for that side.
  • AGENTS.md pointer. Each skill gets a ### Title entry under ## Skills with a one-paragraph pointer to the SKILL.md path.
  • Create or merge. Add a new skill when a substantial repository concern would otherwise inflate AGENTS.md beyond an index. Merge sibling concerns into one multi-section skill when they share most of their structure (multi-agent/ is the precedent).
  • Headings are plain. No chapter numbers in skill or AGENTS.md headings. Use descriptive titles.