Setup Command Reference

Last reviewed: 2026-05-08

Freshness source: cmd/bd/setup*.go and internal/recipes/.

For: Setting up beads integration with AI coding tools Version: current CLI behaviour; verify recipe lists against the freshness source above.

Overview

The bd setup command uses a recipe-based architecture to configure beads integration with AI coding tools. Recipes define where workflow instructions are written—built-in recipes handle popular tools, and you can add custom recipes for any tool.

bd prime as SSOT

bd prime is the single source of truth for operational workflow commands. The beads section in each tool's instruction file provides a pointer to bd prime for hook-enabled agents (Claude, Gemini) or the full command reference for AGENTS-first agents (Factory, Mux). Codex uses a Beads skill, generated AGENTS.md guidance, and native Codex hooks.

Template Profiles

Each integration uses one of two profiles that control how much content is written to tool instruction files (AGENTS.md, CLAUDE.md, GEMINI.md, or .github/copilot-instructions.md):

Hook-enabled agents (Claude, Copilot CLI, Gemini) use the minimal profile because bd prime injects full context at session start. AGENTS-first agents use the full profile because their instruction file remains the primary integration surface. Skill-aware agents use .agents/skills/beads/SKILL.md, with project AGENTS.md or global $CODEX_HOME/AGENTS.md/~/.codex/AGENTS.md telling Codex when to use the skill.

Profile precedence: If a file already has a full profile section and a minimal profile tool installs to the same file (e.g., via symlinks), the full profile is preserved to avoid information loss.

Policy Profiles

Template profiles control how much text gets installed. Policy profiles control what an agent is authorized to do at handoff:

The generated Beads block and bd prime default to conservative git authority. Projects that want team-maintainer behavior should say so in their own top-level instructions; Beads does not infer that authority merely because a remote exists.

Built-in Recipes

Quick Start

# List all available recipes
bd setup --list

# Install integration for your tool
bd setup cursor     # Cursor IDE
bd setup windsurf   # Windsurf
bd setup kilocode   # Kilo Code
bd setup claude     # Claude Code
bd setup copilot    # GitHub Copilot CLI plugin + instructions
bd setup gemini     # Gemini CLI
bd setup factory    # Factory.ai Droid
bd setup codex      # Beads agent skill + AGENTS.md guidance + native hooks
bd setup mux        # Mux
bd setup aider      # Aider

# Verify installation
bd setup cursor --check
bd setup claude --check

# Print template to stdout (for inspection)
bd setup --print

# Write template to custom path
bd setup -o .my-editor/rules.md

# Add a custom recipe
bd setup --add myeditor .myeditor/rules.md
bd setup myeditor  # Now you can use it

Factory.ai (Droid)

Factory.ai Droid integration uses the AGENTS.md standard, which is compatible with multiple AI coding assistants.

Installation

# Create or update AGENTS.md with beads integration
bd setup factory

What Gets Installed

Creates or updates AGENTS.md in your project root with:

• Issue tracking workflow instructions
• Quick command reference
• Issue types and priorities
• Dolt remote sync explanation
• Important rules for AI agents

The beads section is wrapped in HTML comments (<!-- BEGIN/END BEADS INTEGRATION -->) with metadata for safe updates. The begin marker includes profile and hash metadata (e.g., <!-- BEGIN BEADS INTEGRATION profile:full hash:d4f96305 -->) for freshness detection. Legacy markers without metadata are auto-upgraded on the next install or update.

AGENTS.md Standard

AGENTS.md is an industry-standard format for AI coding agent instructions, supported by:

• Factory.ai Droid - Specialized coding agents
• Cursor - Also reads AGENTS.md (in addition to .cursor/rules)
• Aider - Can be configured to read AGENTS.md
• Gemini CLI - Google's command-line AI assistant
• Jules - Google's coding assistant
• Codex - OpenAI's code generation model
• Zed - AI-enhanced editor
• And many more emerging tools

Using AGENTS.md means one configuration file works across your entire AI tool ecosystem.

Flags

Examples

# Check if beads section is in AGENTS.md
bd setup factory --check
# Output: ✓ Factory.ai integration installed: AGENTS.md
#         Beads section found in AGENTS.md

# Remove beads section
bd setup factory --remove

How It Works

Factory Droid and other AGENTS.md-compatible tools automatically read AGENTS.md from:

1. Current working directory (./AGENTS.md)
2. Parent directories up to repo root
3. Personal override (~/.factory/AGENTS.md)

The beads section teaches AI agents:

• To use bd ready for finding work
• To use bd create for tracking new issues
• To treat commit, push, and Dolt remote sync as policy-controlled handoff actions
• The complete workflow pattern and best practices

Updating Existing AGENTS.md

If you already have an AGENTS.md file with other project instructions:

• bd setup factory will append the beads section
• Re-running it will update the existing beads section (idempotent)
• The markers (<!-- BEGIN/END BEADS INTEGRATION -->) ensure safe updates

When to Use This vs Other Integrations

Use Factory integration when:

• ✅ You use Factory.ai Droid
• ✅ You want one config file for multiple AI tools
• ✅ You prefer the AGENTS.md standard
• ✅ Your team uses multiple AI coding assistants

Use other integrations when:

• ✅ You only use Claude Code → bd setup claude (hooks are more dynamic)
• ✅ You need tool-specific features (like Claude's stealth mode)

You can use multiple integrations simultaneously - they complement each other!

Codex CLI

Codex reads project instructions from AGENTS.md in the current working directory or project root, and global instructions from $CODEX_HOME/AGENTS.md when CODEX_HOME is set, otherwise ~/.codex/AGENTS.md. The Codex setup path installs the generic beads agent skill, writes managed Codex guidance, enables native hooks, and installs a fallback hooks file when the Beads Codex plugin is not managing hooks.

Installation

bd setup codex          # Project Beads skill + AGENTS.md guidance + native hooks
bd setup codex --global # Global Beads skill + guidance + native hooks

What Gets Installed

Project install (bd setup codex):

• Creates or updates .agents/skills/beads/SKILL.md
• Creates or updates .agents/skills/beads/agents/openai.yaml
• Creates or updates project AGENTS.md with a marked section generated by bd setup codex
• Creates or updates .codex/config.toml with native hooks enabled
• Creates or updates .codex/hooks.json unless hooks are plugin-managed

Global install (bd setup codex --global):

• Creates or updates ~/.agents/skills/beads/SKILL.md
• Creates or updates ~/.agents/skills/beads/agents/openai.yaml
• Creates or updates $CODEX_HOME/AGENTS.md when CODEX_HOME is set, otherwise ~/.codex/AGENTS.md, with a marked section generated by bd setup codex --global
• Creates or updates global Codex hook configuration under $CODEX_HOME when set, otherwise ~/.codex

Flags

Notes

• Restart Codex if it's already running to pick up the new instructions.
• The plugin package under plugins/beads/ is separate from bd setup codex. The setup command writes a small setup-only skill and managed guidance into the target repository or user-level .agents directory.
• bd init runs project bd setup codex automatically unless --skip-agents or --stealth is used.
• In worktree/shared/BEADS_DIR setups, use bd where to confirm the resolved workspace; the integration does not require a local ./.beads.
• bd setup codex uses its own marker pair (BEGIN/END BEADS CODEX SETUP), distinct from the BEGIN/END BEADS INTEGRATION markers used by bd setup factory and bd setup mux. Running both bd setup codex and bd setup factory/mux against the same AGENTS.md will leave two managed sections side by side; each bd setup … --check only inspects its own section, and bd setup … --remove only removes its own section.

Mux

Mux reads layered instruction files, including workspace AGENTS.md. Adding the beads section is enough to get Mux and beads working together.

Installation

bd setup mux            # Root AGENTS.md
bd setup mux --project  # Root AGENTS.md + .mux/AGENTS.md
bd setup mux --global   # Root AGENTS.md + ~/.mux/AGENTS.md

What Gets Installed

Creates or updates AGENTS.md with the beads integration section (same markers as Factory.ai and Codex).

Notes

• Mux instruction file behavior is documented at https://mux.coder.com/AGENTS.md.
• Restart the workspace session if Mux is already running.

Flags

Claude Code

Claude Code integration uses hooks to automatically inject beads workflow context at session start and before context compaction.

Installation

# Global installation (recommended)
bd setup claude

# Project-only installation
bd setup claude --project

# With stealth mode (flush only, no git operations)
bd setup claude --stealth

What Gets Installed

Global installation (~/.claude/settings.json):

• SessionStart hook: Runs bd prime --hook-json when a session starts, resumes, clears, or restarts after compaction

Project installation (.claude/settings.local.json):

• Same hooks, but only active for this project

Instruction file (CLAUDE.md in project root):

• Minimal-profile beads section pointing to bd prime
• Managed with hash/version markers for safe updates and --check freshness detection

Flags

Examples

# Check hooks + CLAUDE.md beads section
bd setup claude --check
# Output: ✓ Global hooks installed: /Users/you/.claude/settings.json
#         ✓ Claude Code integration installed: /path/to/CLAUDE.md (current)

# Remove hooks
bd setup claude --remove

# Install project-specific hooks with stealth mode
bd setup claude --project --stealth

How It Works

The hook calls bd prime --hook-json which:

1. Outputs workflow context wrapped in the SessionStart JSON envelope Claude Code expects
2. Prints persistent memories near the top so hook-output previews do not hide them
3. Starts with a truncation warning telling agents to read the full persisted hook output when the host caps previews
4. Ensures Claude always knows how to use beads
5. Follows resolved workspace semantics, so bd where is the right diagnostic check when local ./.beads is absent

For low-token hooks that only need durable project facts, use bd prime --memories-only.

This is more context-efficient than MCP tools (~1-2k tokens vs 10-50k for MCP schemas).

Gemini CLI

Gemini CLI integration uses a SessionStart hook to automatically inject beads workflow context when a session opens.

Installation

# Global installation (recommended)
bd setup gemini

# Project-only installation
bd setup gemini --project

# With stealth mode (flush only, no git operations)
bd setup gemini --stealth

What Gets Installed

Global installation (~/.gemini/settings.json):

• SessionStart hook: Runs bd prime --hook-json when a new session starts, wrapped in the JSON envelope Gemini's hook contract requires

Project installation (.gemini/settings.json):

• Same hooks, but only active for this project

Instruction file (GEMINI.md in project root):

• Minimal-profile beads section pointing to bd prime
• Managed with hash/version markers for safe updates and --check freshness detection

Flags

Examples

# Check hooks + GEMINI.md beads section
bd setup gemini --check
# Output: ✓ Global hooks installed: /Users/you/.gemini/settings.json
#         ✓ Gemini CLI integration installed: /path/to/GEMINI.md (current)

# Remove hooks
bd setup gemini --remove

# Install project-specific hooks with stealth mode
bd setup gemini --project --stealth

How It Works

The hooks call bd prime which:

1. Outputs workflow context for Gemini to read
2. Prints persistent memories near the top so hook-output previews do not hide them
3. Starts with a truncation warning telling agents to read the full persisted hook output when the host caps previews
4. Ensures Gemini always knows how to use beads

For low-token hooks that only need durable project facts, use bd prime --memories-only.

This works similarly to Claude Code integration, using Gemini CLI's hook system (SessionStart event). Unlike Claude Code, Gemini requires hook stdout to be valid JSON — bd prime --hook-json wraps the markdown in the required envelope.

Cursor IDE

Cursor integration creates a rules file that provides beads workflow context to the AI.

Installation

bd setup cursor

What Gets Installed

Creates .cursor/rules/beads.mdc with:

• Core workflow rules (track work in bd, not markdown TODOs)
• Quick command reference
• Workflow pattern (ready → claim → work → close → sync)
• Context loading instructions

Flags

Examples

# Check if rules are installed
bd setup cursor --check
# Output: ✓ Cursor integration installed: .cursor/rules/beads.mdc

# Remove rules
bd setup cursor --remove

How It Works

Cursor reads .cursor/rules/*.mdc files and includes them in the AI's context. The beads rules file teaches the AI:

• To use bd ready for finding work
• To use bd create for tracking new issues
• To treat commit, push, and Dolt remote sync as policy-controlled handoff actions
• The basic workflow pattern

Aider

Aider integration creates configuration files that teach the AI about beads, while respecting Aider's human-in-the-loop design.

Installation

bd setup aider

What Gets Installed

Flags

Examples

# Check if config is installed
bd setup aider --check
# Output: ✓ Aider integration installed: .aider.conf.yml

# Remove configuration
bd setup aider --remove

How It Works

Unlike Claude Code, Aider requires explicit command execution. The AI will suggest bd commands, which the user runs via /run:

You: What issues are ready to work on?

Aider: Let me check. Run:
/run bd ready

You: [runs the command]

Aider: Great! To claim bd-42, run:
/run bd update bd-42 --claim

This respects Aider's philosophy of keeping humans in control while still leveraging beads for issue tracking.

Comparison

Best Practices

1. Start with Factory integration - Creates AGENTS.md which works across multiple AI tools:

   bashCopy

   bd setup factory
   

2. Add tool-specific integrations as needed - Claude hooks, Cursor rules, or Aider config for tool-specific features

3. Install globally for Claude Code or Gemini CLI - You'll get beads context in every project automatically

4. Use stealth mode in CI/CD - bd setup claude --stealth or bd setup gemini --stealth avoids git operations that might fail in automated environments

5. Commit instruction files to git - This ensures all team members and AI tools get the same instructions (AGENTS.md, CLAUDE.md, GEMINI.md, as applicable)

6. Run bd doctor after setup - Verifies the integration is working:

   bashCopy

   bd doctor | grep -iE "claude|gemini"
   # Claude Integration: Hooks installed (CLI mode)
   # Gemini CLI Integration: Hooks installed
   

Troubleshooting

"Hooks not working"

1. Restart your AI tool after installation
2. Verify with bd setup <tool> --check
3. Check bd doctor output for integration status

"Context not appearing"

For Claude Code, ensure bd prime works standalone:

bd prime

If this fails, fix the underlying beads issue first.

"Want to switch from global to project hooks"

# Remove global hooks
bd setup claude --remove

# Install project hooks
bd setup claude --project

Custom Recipes

You can add custom recipes for editors/tools not included in the built-in list.

Adding a Custom Recipe

# Add a recipe that writes to a specific path
bd setup --add myeditor .myeditor/rules.md

# Install it
bd setup myeditor

# Check it
bd setup myeditor --check

# Remove it
bd setup myeditor --remove

User Recipes File

Custom recipes are stored in .beads/recipes.toml:

[recipes.myeditor]
name = "myeditor"
path = ".myeditor/rules.md"
type = "file"

Using Arbitrary Paths

For one-off installs without saving a recipe:

# Write template to any path
bd setup -o .my-custom-location/beads.md

# Inspect the template first
bd setup --print

Recipe Types

Custom recipes added via --add are always type file.

Related Documentation

• CLAUDE_INTEGRATION.md - Design decisions for Claude Code integration
• AIDER_INTEGRATION.md - Detailed Aider workflow guide
• QUICKSTART.md - Getting started with beads
• CLI_REFERENCE.md - Full command reference