Agent Assistant

One AI. A Full Engineering Team. Ship in Hours.

Turn your Claude Code, Cursor, GitHub Copilot, Antigravity, Codex, Kiro, or Qwen into 21 specialist agents — backend engineer, frontend engineer, QA, security auditor, database architect, and 16 more — that work together through structured workflows with parallel execution, adversarial review, and automatic skill injection.

85% fewer tokens. 70% fewer bugs. Features shipped in hours, not days.

npm i -g @namch/agent-assistant && agent-assistant install --all

Works with Claude Code · Cursor · GitHub Copilot · Codex · Antigravity · Kiro · Qwen

See It in Action

Before Agent Assistant: You prompt your AI → it writes code → it misses edge cases → you fix bugs → you write docs yourself → you run tests manually → 3 days pass.

After Agent Assistant: You type /cook:hard "add Stripe billing with usage-based pricing" → a tech lead plans, a backend engineer implements, a reviewer audits, and a tester validates → in 90 minutes, every file is written, tested, and documented.

The Numbers That Actually Matter

| Metric | Without Agent Assistant | With Agent Assistant | Improvement | |--------|------------------------|----------------------|-------------| | Time to ship a feature | 3 days | 4 hours | 70% faster | | Bug rate per feature | ~12 issues | ~3 issues | 75% fewer bugs | | AI token consumption | Full project in context | Bounded, targeted calls | 85% fewer tokens | | Documentation coverage | None (or stale) | Auto-generated, always current | 100% maintained |

The Killer Feature: Run /wiki on Any Codebase

One command. Wake up to a complete, cross-linked, searchable knowledge base.

/wiki:team document my-codebase
# → 7-phase adversarial review begins
# → 30+ wiki pages generated: entities, concepts, decisions, runbooks
# → Auto-generated knowledge graph: 42 nodes, ~175 cross-links
# → Query with: /wiki query "How does auth work?"

The wiki isn't just documentation. Every page is generated by one AI, reviewed by a second, and challenged by a third (the Golden Triangle). Run it on your project right now:

/wiki:fast                    # Bootstrap — 5 essential pages in 3 min
/wiki:hard "full codebase"    # Comprehensive — 25 pages with review
/wiki:team "mission-critical"  # Adversarial — 30+ pages, 7 phases

• 9 page types: Summaries, Entities, Concepts, Decisions, Chronicles, Runbooks, Comparisons, Syntheses, Postmortems
• Always current: Re-run compile after any refactor — the wiki regenerates from live code
• Query instead of search: Ask natural language questions, get answers with verified source citations

Every Command is a Team

| You type | What happens | |----------|-------------| | /cook:fast "add dark mode" | 2 agents sprint a small feature | | /cook:hard "implement OAuth 2.0" | 5-8 agents with full quality gates | | /cook:team "mission-critical feature" | 10+ agents with Golden Triangle | | /fix "payment fails on Safari" | Debugger traces → engineer patches → reviewer validates | | /test:hard "user registration flow" | QA engineer writes integration + unit tests | | /review "audit the auth module" | Security engineer + code reviewer audit | | /docs:core | Generates technical docs from your codebase | | /plan "build notification system" | Planner + architect produce a full RFC | | /deploy:production | DevOps engineer handles build, deploy, rollback |

Each command has 3 quality tiers:

| Variant | When to use | Agents deployed | |---------|-------------|----------------| | :fast | Simple, low-risk | 2-3 agents | | :hard | Complex, production code | 5-8 agents + quality gates | | :team | Mission-critical, full scrutiny | 10+ agents + Golden Triangle |

The Golden Triangle — Adversarial Quality

For :team workflows, every output passes through three agents who debate, not just approve:

graph TB
    TL["👑 Tech Lead<br/>Plans · Decomposes · Arbitrates"]
    EX["🔍 Executor<br/>Writes · Builds · Defends"]
    RV["🛡️ Reviewer<br/>Challenges · Validates · Blocks"]

    TL -->|"decomposes"| EX
    TL -->|"decomposes"| RV
    EX -->|"proposes"| RV
    RV -->|"challenges"| EX
    EX -->|"defends or revises"| RV
    TL -->|"arbitrates disputes"| RV
    RV -->|"approved"| TL
    TL -->|"consensus reached"| DONE["✅ Ship"]

    style TL fill:#7c3aed,color:#fff
    style EX fill:#059669,color:#fff
    style RV fill:#ea580c,color:#fff

No agent can slide a bad decision through. The executor implements, the reviewer attacks, and the tech lead arbitrates until consensus is reached.

How We Compare

| Feature | Agent Assistant | Cursor Composer | Copilot Agents | Claude Code | |---------|-----------------|-----------------|----------------|-------------| | Agents | 21 specialist | 1 unified | 1 unified | CLI only | | Adversarial Review | ✅ Golden Triangle | ❌ | ❌ | ❌ | | Auto Wiki | ✅ /wiki command | ❌ | ❌ | ❌ | | Skill Matrix | 78 skills | Built-in | Built-in | Built-in | | Multi-Tool Support | 7 tools | Cursor only | VS Code only | Claude only | | CLI-First | ✅ | ❌ | ❌ | ✅ | | Variants (fast/hard/team) | ✅ | ❌ | ❌ | ❌ | | Knowledge Graph | ✅ Auto-generated | ❌ | ❌ | ❌ | | Matrix Skills | ✅ Community extensibility | ❌ | ❌ | ❌ |

Agent Assistant enhances your existing AI coding tool — it doesn't replace it.

21 Specialist Agents

| Domain | Agents | What they own | |--------|--------|---------------| | Implementation | backend-engineer, frontend-engineer, mobile-engineer, game-engineer | All production code | | Architecture | tech-lead, database-architect | System design, schema, trade-offs | | Quality | tester, reviewer, debugger, security-engineer | All quality gates | | Planning | planner, brainstormer, business-analyst | Strategy, estimation, requirements | | Support | designer, devops-engineer, docs-manager, performance-engineer, researcher, scouter, project-manager, reporter | Everything else |

Matrix Skill Discovery — 78+ Skills

Agents don't ship with hardcoded skills. They declare a profile, and the Matrix auto-injects the right skills at runtime:

# Agent declares:
profile: "backend:execution"

# Matrix resolves → 20+ skills auto-injected:
# FastAPI, Prisma, Postgres, Redis, Docker, Kubernetes,
# JWT, OAuth2, rate limiting, error handling, testing...

78+ skills across 19 domains. A skill added to the Matrix is instantly available to every relevant agent. No manual config.

Quick Start

1. Install (2 minutes)

# Install the package
npm install -g @namch/agent-assistant@latest

# Set up for your AI tool
agent-assistant install cursor        # or claude, copilot, antigravity, codex, kiro, qwen
agent-assistant install --all        # install for every tool you use

2. Try the wiki on this codebase (recommended first step)

/wiki:fast                    # Bootstrap — 5 essential pages in 3 min
/wiki:hard "full codebase"    # Comprehensive — 25 pages with review
/wiki:team "mission-critical"  # Adversarial — 30+ pages, 7 phases

3. Build anything

/cook:fast "add dark mode toggle"
/cook:hard "implement Stripe billing"
/fix "checkout button broken on mobile"
/test:hard "payment flow"
/review "API authentication layer"

Project Structure

agent-assistant/
├── agents/          # 21 specialist agents (markdown definitions)
├── commands/        # 50+ workflow commands (fast/hard/team variants)
├── rules/           # 8 orchestration rules (CORE, PHASES, AGENTS, TEAMS...)
├── matrix-skills/   # 19 domain skill registries
├── skills/         # 78+ domain skills
├── wiki/           # Built-in wiki documentation (try /wiki:team here!)
└── cli/             # Installer (installs into ~/.cursor/, ~/.claude/, etc.)

Supported Tools

| Tool | Status | Install Path | |------|--------|-------------| | Cursor | Full support | ~/.cursor/ | | Claude Code | Full support | ~/.claude/ | | GitHub Copilot | Full support | ~/.copilot/ | | Antigravity/Gemini | Full support | ~/.antigravity/ + ~/.gemini/ | | Codex | Full support | ~/.codex/ | | Kiro | Full support | ~/.kiro/ | | Qwen | Full support | ~/.qwen/ |

Contributing

We're building the multi-agent orchestration standard. Here's how you can contribute:

| What to contribute | How to start | Time needed | |--------------------|--------------|-------------| | Add a Matrix Skill | Drop a skill file into skills/<domain>/ — see skills/README.md for the template | ~10 min | | Create a new Agent | Define an agent in agents/ with a profile and responsibilities | ~20 min | | Add a Command variant | Extend an existing command with a new :variant | ~30 min | | Improve documentation | Every wiki page and README section is fair game | ~5 min |

See CONTRIBUTING.md for the full guide.

Roadmap

• [ ] v2.0 — Web dashboard for wiki visualization and team monitoring
• [ ] v2.1 — VS Code extension — full IDE integration, not just CLI
• [ ] v2.2 — Agent marketplace — publish and install community agents
• [ ] v2.3 — Real-time collaboration mode — multiple agents on the same PR
• [ ] v2.4 — GitHub App — auto-comment PR reviews and wiki generation on repo events

Want to influence the roadmap? Open an issue, start a discussion, or submit a PR.

Support

If this helps you ship faster, consider buying me a coffee!

License

MIT — NamCH — Issues — Discussions

Agent Assistant — One AI. A whole team. Zero compromise on quality.