Getting Started with Agenthood

A full AI engineering team — install in five minutes, invoke on demand.

What is Agenthood

Agenthood is a suite of specialized AI agents that automate the work around your code: commit messages, PR reviews, security audits, changelogs, and more. Two modes:

Runtime mode (flagship) — agents run autonomously via the TypeScript CLI. They reason, use tools, and produce results without manual prompting.
Skill mode (alternative) — members load as Markdown skill files into Claude Code, Copilot, or any agent runtime that supports skill files.

Both modes use the same member definitions. Choose runtime for automation, skill mode for assisted workflows.

Quick install


npm install --save-dev agenthood
npx agenthood init          # interactive setup (~2 minutes)
npx agenthood check         # verify everything is in place

Requirements: Node.js 22+, git, and gh CLI for PR sync. No API key required for basic setup.

Essential commands

Command	What it does
`init`	Install hooks, templates, and member skills into your project
`check`	Health check — confirm the Society is fully operational
`run <member> "<task>"`	Invoke a member as an autonomous agent
`list`	List all members with activation status
`pr-sync --pr <N>`	Auto-sync PR body with new commits
`activate <member>`	Enable a member's skill file in your runtime
`deactivate <member>`	Disable a member's skill file
`oath`	Read the Society oath
`eject`	Remove all Society files from your project
`diff [member]`	Show SKILL.md changes vs the versions locked in `agenthood.lock`

Your first commit workflow

Write code — make changes in your project.
Stage — git add -p
Commit — git commit. The Doorman validates your message against conventional commits. If it fails, you get a clear reason and a suggestion.
Push — pre-push hooks run tests before the branch leaves your machine.
Open a PR — the PR body is pre-filled with the Society's template (## What changed, ## Why, ## How to test).
Review — The Reviewer checks correctness, security, performance, maintainability, and test coverage.
Release — The Herald determines the semver bump, generates the changelog, and publishes to npm.

Every step is automated. Every step has a member responsible.

CI pipeline

The Society ships 7 workflows that enforce standards on every push and PR:

Workflow	Members	What it does
`pr.yml`	Doorman, Auditor, Scribe, Warden, Librarian, Sentinel, Tester	7 merged PR checks: commit messages, secrets, labels, file sizes, docs sync, member structure, test suite
`reviewer.yml`	The Reviewer	Reviews every pushed diff via LLM, posts findings as PR comment
`herald.yml`	The Herald	Posts a CI summary comment on every PR
`semantic-release.yml`	The Herald	Automated release and npm publish
`vscode-extension.yml`	The Envoy	Builds, tests, and packages the VS Code extension
`publish-vsce.yml`	The Envoy	Publishes extension to VS Code Marketplace on release
`distribution.yml`	The Envoy	Skills.sh verification, SkillsMP schema validation, traction badges

These run on GitHub Actions. Every check must pass before merge.

PR body sync

The Reviewer keeps PR descriptions accurate as new commits land (replaced The Manuscript in v2.0.0). A  marker splits the body into two zones: your narrative above, auto-generated content below.


# Fast path — no API key required, runs in CI
npx agenthood pr-sync --pr 42

# Preview changes without publishing
npx agenthood pr-sync --pr 42 --dry-run

# Context-aware — The Scribe loads your conventions and ADRs first
npx agenthood run the-scribe "sync PR #42"

On each run, The Scribe detects new commits since the last sync, updates the ## What Changed section, and posts a reviewer comment. Your ## Why and ## How to Test sections are never touched.

For the full breakdown, see the Skills Reference.

Configuration

The Society reads from .agenthood/config.json, scaffolded by init:


{
  "version": "1",
  "runtime": "claude-code",
  "members": ["the-scribe", "the-architect", "the-reviewer", "..."],
  "hooks": { "hooksPath": ".githooks" },
  "conventions": {
    "commitTemplate": ".gitmessage",
    "commitlintConfig": "commitlint.config.ts"
  },
  "providers": [
    { "name": "opencode", "model": "mimo-v2.5", "priority": 1 },
    { "name": "opencode-go", "model": "mimo-v2.5", "priority": 2 },
    { "name": "anthropic", "model": "claude-sonnet-5", "priority": 3 }
  ],
  "qualityGates": { "typescript": true, "tests": true, "lint": true }
}

Key environment variables:

The runtime automatically loads variables from a .env file in your project root.

Variable	Required for	Default
`GROQ_API_KEY`	Runtime mode fallback (free at console.groq.com)	—
`OPENCODE_API_KEY`	Runtime mode default provider (opencode, free tier at opencode.ai)	—
`ANTHROPIC_API_KEY`	Runtime mode, high-complexity tasks	—
`OPENAI_API_KEY`	Runtime mode (fallback provider)	—
`GITHUB_TOKEN`	PR sync (auto-set in CI)	—

No API key is needed for skill-file mode or for pr-sync in CI.

Alternative path: skill files

If you use Claude Code, Copilot, or another assistant, load members as skill files:


npx agenthood init
# Select your runtime when prompted — skills are installed automatically

The assistant loads the member's SKILL.md as context. Ask the Reviewer to check your PR, ask the Auditor to scan your auth flow. They know their role.

Next steps

5-minute skill creation quickstart — one working SKILL.md in under 20 lines
Skills Reference — all members, their tools, and invocation syntax
Level 1: GenAI & RAG Basics — LLMs, prompt engineering, RAG
Level 2: AI Agent Essentials — memory, planning, multi-agent
Level 3: Advanced Agent Skills — integration, performance, production
Architecture Decision Records — why the Society is built the way it is
Online docs — hosted version of this guide

Install in five minutes. The standards last forever.