Create Beads Orchestration
Set up lightweight multi-agent orchestration with git-native task tracking and mandatory code review gates.
CRITICAL: Mandatory 4-Step Workflow
<mandatory-workflow> You MUST follow ALL 4 steps below in exact order. Missing ANY step is a CATASTROPHIC FAILURE.| Step | Action | Checkpoint | |------|--------|------------| | 1 | Get project info from user | Have project name, directory, AND provider choice | | 2 | Clone repo and run bootstrap | Bootstrap completes successfully | | 3 | STOP - Instruct user to restart Claude Code | User confirms they will restart | | 4 | After restart: Run discovery agent | Supervisors created in .claude/agents/ |
DO NOT:
- Skip asking for project info
- Skip asking about provider delegation (Claude-only vs External providers)
- Continue after bootstrap without telling user to restart
- Forget to run discovery after restart
- Consider setup complete until discovery has run
The setup is NOT complete until Step 4 (discovery) has run. </mandatory-workflow>
Step 1: Get Project Info
<critical-step1> **YOU MUST ASK ALL THREE QUESTIONS BEFORE PROCEEDING TO STEP 2 using AskUserQuestion.**- Project directory: Where to install (default: current working directory)
- Project name: For agent templates (will auto-infer from package.json/pyproject.toml if not provided)
- Provider delegation: MANDATORY - You MUST use AskUserQuestion for this choice </critical-step1>
1.1 Get Project Directory and Name
Ask the user or auto-detect from package.json/pyproject.toml.
1.2 MANDATORY: Ask Provider Delegation Choice
<mandatory-question> **YOU MUST CALL AskUserQuestion WITH THIS EXACT QUESTION BEFORE RUNNING BOOTSTRAP.**Do NOT skip this. Do NOT assume a default. Do NOT proceed without the user's explicit choice.
AskUserQuestion(
questions=[{
"question": "How should read-only agents (scout, detective, architect, scribe, code-reviewer) be executed?",
"header": "Providers",
"options": [
{"label": "Claude only (Recommended)", "description": "All agents run via Claude Task(). Simpler setup, no external dependencies."},
{"label": "External providers", "description": "Delegate to Codex CLI (with Gemini fallback). Requires codex login and optional gemini CLI."}
],
"multiSelect": false
}]
)
After user answers:
- If "Claude only" → use
--claude-onlyflag in bootstrap - If "External providers" → do NOT use
--claude-onlyflag </mandatory-question>
DO NOT proceed to Step 2 until you have the provider choice from the user.
Step 2: Clone and Run Bootstrap
git clone --depth=1 https://github.com/AvivK5498/The-Claude-Protocol "${TMPDIR:-/tmp}/beads-orchestration-setup"
# If user selected "Claude only":
python3 "${TMPDIR:-/tmp}/beads-orchestration-setup/bootstrap.py" \
--project-name "{{PROJECT_NAME}}" \
--project-dir "{{PROJECT_DIR}}" \
--claude-only
# If user selected "External providers":
python3 "${TMPDIR:-/tmp}/beads-orchestration-setup/bootstrap.py" \
--project-name "{{PROJECT_NAME}}" \
--project-dir "{{PROJECT_DIR}}"
The bootstrap script will:
- Install beads CLI (via brew, npm, or go)
- Initialize
.beads/directory - Copy agent templates to
.claude/agents/ - Copy hooks to
.claude/hooks/ - Configure
.claude/settings.json - Set up
.mcp.jsonfor provider_delegator - Create
CLAUDE.mdwith orchestrator instructions - Update
.gitignore
Verify bootstrap completed successfully before proceeding.
Step 3: STOP - User Must Restart
<critical> **YOU MUST STOP HERE AND INSTRUCT THE USER TO RESTART CLAUDE CODE.**Tell the user:
Setup phase complete. You MUST restart Claude Code now.
The new hooks and MCP configuration will only load after restart.
After restarting:
- Open this same project directory
- Tell me "Continue orchestration setup" or run
/create-beads-orchestrationagain- I will run the discovery agent to complete setup
Do not skip this restart - the orchestration will not work without it.
DO NOT proceed to Step 4 in this session. The restart is mandatory. </critical>
Step 4: Run Discovery (After Restart)
<post-restart> If the user returns after restart and says "continue setup" or similar:- Verify bootstrap completed (check for
.claude/agents/scout.md) - Run the discovery agent:
Task(
subagent_type="discovery",
prompt="Detect tech stack and create supervisors for this project"
)
Discovery will:
- Scan package.json, requirements.txt, Dockerfile, etc.
- Fetch specialist agents from external directory
- Inject beads workflow into each supervisor
- Write supervisors to
.claude/agents/
- After discovery completes, tell the user:
Orchestration setup complete!
Created supervisors: [list what discovery created]
You can now use the orchestration workflow:
- Create tasks with
bd create "Task name" -d "Description"- The orchestrator will delegate to appropriate supervisors
- All work requires code review before completion </post-restart>
Cleanup (Optional)
rm -rf "${TMPDIR:-/tmp}/beads-orchestration-setup"
What This Creates
- Beads CLI for git-native task tracking (one bead = one branch = one task)
- Core agents: scout, detective, architect, scribe, code-reviewer
- Discovery agent: Auto-detects tech stack and creates specialized supervisors
- Hooks: Enforce orchestrator discipline, code review gates, concise responses
- Branch-per-task workflow: Parallel development with automated merge conflict handling
With --claude-only (default):
- All agents run via Claude Task() - no external dependencies
With external providers:
- MCP Provider Delegator enables Codex→Gemini→Claude fallback chain
- Additional enforcement hooks for provider delegation
Requirements
Claude only mode (default):
- beads CLI: Installed automatically (or manually via brew/npm/go)
- uv: Python package manager (only if using external providers)
External providers mode:
- Codex CLI:
codex loginfor authentication (primary provider) - Gemini CLI: Optional fallback when Codex hits rate limits
- uv: Python package manager for MCP server
More Information
See the full documentation: https://github.com/AvivK5498/The-Claude-Protocol