Skip to content

0xranx/agentbrief

Repository files navigation

AgentBrief

CI npm License: MIT

One command turns your AI coding agent into a domain specialist.

Website · Catalog · npm · 中文

Your Claude Code, Cursor, OpenCode, or Codex gives generic answers because it doesn't know your domain. AgentBrief fixes that — install a brief and your agent gains real expertise: security auditing, code review, product specs, growth hacking, and more.

npx agentbrief use fullstack-engineer
# Your agent now enforces strict TypeScript, follows Next.js conventions,
# builds accessible UIs, and reviews PRs like a staff engineer — 9 skills in 1 command

Before vs After

WITHOUT a brief:
  You: "Review this code for issues"
  Agent: "The code looks good overall. Consider adding error handling."

WITH security-auditor brief:
  You: "Review this code for issues"
  Agent: "CWE-89 CRITICAL at line 23: SQL injection via string concatenation.
         Attack vector: attacker injects arbitrary SQL through userId param.
         Fix: const query = 'SELECT * FROM users WHERE id = $1';
              await db.query(query, [userId]);"

Install

npm install -g agentbrief
# or
pnpm add -g agentbrief

Quick Start Packs

Not sure where to begin? Pick a combo — each one bundles multiple specialist briefs into a single command:

  fullstack-engineer  →  release-engineer  →  startup-founder
       Build                 Ship                 Grow
Pack What your agent gains
fullstack-engineer Strict TypeScript + Next.js + accessible UI + PR reviews (9 skills)
release-engineer QA testing + security review + CI/CD + documentation (10 skills)
startup-founder Product specs + SEO + growth analytics + security + launch (12 skills)
agentbrief use fullstack-engineer   # Build: write production code
agentbrief use release-engineer     # Ship: test, secure, deploy, document
agentbrief use startup-founder      # Grow: product, growth, launch strategy

All Briefs

Code Quality & Engineering:

Brief Your agent becomes...
security-auditor OWASP security reviewer who cites CWE numbers
code-reviewer Staff engineer who catches architecture + logic issues
qa-engineer QA who finds bugs, writes tests, fixes with atomic commits
typescript-engineer Type safety enforcer — zero any, exhaustive checks
nextjs-developer Next.js 15 specialist (App Router, RSC, Tailwind)
design-engineer Design engineer with 80-item review checklist
devops-sre SRE who sets up CI/CD, monitoring, incident response
tech-writer Documentation specialist with API docs + release notes

Product, Growth & Business:

Brief Your agent becomes...
product-manager PM who writes PRDs with RICE/ICE prioritization
growth-engineer Growth hacker with SEO audit + analytics + content strategy
data-analyst BI analyst with metrics frameworks + SQL patterns
startup-advisor Startup advisor with CEO review + launch planning
social-media-manager Social media manager — Twitter/X + Xiaohongshu + Douyin research
feishu-writer Feishu/Lark specialist — docs, wikis, messaging, tasks (official lark-cli)

Browse the full Catalog or the Website for details on each brief.

Usage

# Apply from the official registry
agentbrief use security-auditor

# Apply from GitHub
agentbrief use github:owner/repo
agentbrief use github:owner/[email protected]

# Apply from local path
agentbrief use ./path/to/brief

# Browse, inspect, manage
agentbrief search                 # List all briefs
agentbrief list                   # See what's applied
agentbrief show <name>            # View injected content
agentbrief preview <name>         # Preview without applying
agentbrief update                 # Fetch latest versions
agentbrief eject <name>           # Clean removal

How It Works

AgentBrief compiles a brief (role + knowledge + skills) into the instruction files your AI agent reads, with content optimized per engine:

Engine File Compilation
Claude Code CLAUDE.md Full — personality, knowledge refs, skill triggers
Cursor .cursorrules Minimal — headings + first paragraph + lists
OpenCode AGENTS.md Concise — first sentence per paragraph
Codex AGENTS.md Concise — same as OpenCode

Your existing files are preserved — briefs are injected between <!-- agentbrief:name:start/end --> markers. Eject removes only the brief content.

See what gets injected — example output for fullstack-engineer
<!-- agentbrief:fullstack-engineer:start -->
# AgentBrief: fullstack-engineer

## Role
You are a senior full-stack TypeScript developer. You build production
applications with Next.js 15, React 19, and Tailwind CSS. You enforce
strict type safety and review your own code with principal-engineer rigor.

## Constraints
- Never use `any` — always annotate return types on exports
- Server Components by default — only add 'use client' when needed
- WCAG 2.1 AA minimum — semantic HTML, keyboard navigation

## Skills
- **next-best-practices** — USE WHEN: Writing Next.js code
- **typescript-advanced-types** — USE WHEN: Complex type logic
- **architecture-review** — USE WHEN: Reviewing PRs
- **design-review-checklist** — USE WHEN: Checking UI quality
- **agent-browser** — USE WHEN: Visual verification needed
  ...and 4 more
<!-- agentbrief:fullstack-engineer:end -->

What's a Brief?

my-brief/
├── brief.yaml          # Config: name, version, extends, skills
├── personality.md      # Identity: role, tone, constraints
├── knowledge/          # Reference: domain materials (read on demand)
│   └── cheatsheet.md
└── skills/             # Workflows: executable skill directories
    └── my-skill/
        ├── SKILL.md    # Trigger condition + step-by-step process
        └── ...         # Any supporting files

See briefs/security-auditor/ for a complete example.

Built on Trusted Sources

Every official brief is curated from battle-tested, community-endorsed skills:

Create Your Own

agentbrief init my-agent              # Scaffold
# Edit personality.md, add knowledge/ and skills/
agentbrief use ./my-agent             # Test locally
agentbrief preview ./my-agent         # See compiled output
# Push to GitHub → agentbrief use github:you/my-agent

Want it in the official registry? Submit a PR. Read the Authoring Guide.

Community

Add .agentbrief/ to your .gitignore. Commit the engine files (CLAUDE.md, .cursorrules, AGENTS.md) so your team shares the same agent behavior.

License

MIT

About

Pluggable role definitions for AI coding agents — one command turns Claude Code / Cursor / OpenCode / Codex into a specialized professional

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors