菜单
Create AGENTS.md files with production-ready best practices. Activate when creating new AGENTS.md or implementing quality gates.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Orchestrate static analysis using Codacy. Required PR check on this repo. Use for querying PR analysis, triaging issues, fixing (not suppressing) findings, and local analysis. See SKILL.md for fix patterns and the required-check policy.
Build localized, accessible, premium reader/admin UI with 2026 design standards. Features OKLCH colors, View Transitions, scroll-aware components, and mutual exclusivity panels.
Comprehensive GitHub PR review and automated fix pipeline. Takes a PR number or auto-detects from current branch, runs code review, static analysis, security audit, and quality checks, then produces a structured report and auto-fixes must-fix issues. Activate for "review PR", "fix PR issues", "PR quality check", "review and fix PR #123".
Invoke for complex multi-step tasks requiring intelligent planning and multi-agent coordination. Use when tasks need decomposition, dependency mapping, parallel/sequential/swarm execution strategies, or coordination of multiple specialized agents with quality gates.
Break down complex tasks into atomic, executable goals. Activate for multi-step feature planning, agent coordination, or request decomposition.
Design guidance for AI coding agents. 23 commands, 44 deterministic detector rules, and live browser iteration for AI-generated frontend design. Activate for UI/UX design work, anti-pattern detection, or design quality audits.
| version | 1.0.0 |
| name | agents-md |
| description | Create AGENTS.md files with production-ready best practices. Activate when creating new AGENTS.md or implementing quality gates. |
| category | quality |
| allowed-tools | Read Write Edit Glob |
| license | MIT |
Create comprehensive AGENTS.md files that serve as the single source of truth for AI coding agents.
readonly MAX_LINES_PER_SOURCE_FILE=500
readonly MAX_LINES_PER_SKILL_MD=250
readonly MAX_COMMIT_SUBJECT_LENGTH=72
## TIER 1 — CRITICAL (Must Follow)
- **NEVER commit to `main` directly.** Always use feature branches + PRs.
- **NEVER leak secrets.** Use environment variables, never hardcode.
- **MUST use Argon2id for password hashing.**
| Rule Type | Example |
|---|---|
| Security | Never expose R2 file URLs - use signed URLs |
| Auth | Revoke sessions immediately on permission change |
| Data | Use multi-signal locators for annotations |
## TIER 2 — QUALITY GATES (Blocking)
1. Run `./scripts/quality_gate.sh` before commit.
2. Use atomic commit workflow: `./scripts/atomic-commit/run.sh`
3. Commit messages MUST follow: `type(scope): description`
## TIER 3 — STYLE (Guidelines)
- Max 500 LOC per source file
- No hardcoded environment URLs
- **No hardcoded dates** - Use dynamic dates from environment (e.g., `date +"%Y-%m-%d"`)
- Use Zod for validation, Zustand for state
- Use Vitest + Playwright for testing
## Compliance Self-Check
- [ ] Did I read ALL of AGENTS.md?
- [ ] Did I check Named Constants?
- [ ] Did I verify no secrets in output?
- [ ] Did I run quality gate?
- [ ] Is commit message under 72 chars?
- [ ] Did I use feature branch?
## Skills Reference
| Category | Skills |
|----------|--------|
| Backend | cloudflare-worker-api, secure-invite-and-access |
| Frontend | reader-ui-ux, epub-rendering |
| Testing | testing-strategy, testdata-builders, dogfood |
| Quality | code-quality, security-code-auditor, anti-ai-slop |
# Quality gates
./scripts/quality_gate.sh
# Commit workflow
./scripts/atomic-commit/run.sh --message "type(scope): description" --body "WHY"
- **MUST use CFI locators for EPUB annotations** per ADR-006
- **MUST handle offline-first** with proper sync conflicts
- **MUST validate EPUB structure** before rendering
Good AGENTS.md enables consistent, safe, high-quality AI-assisted development.