Orchestrator Control Skill

Orchestrator Control Skill

This skill spawns a control thread that manages the entire orchestrator, PR shepherds, and multi-agent coordination.

When to Use

Activate this skill when the user wants to:

• Start the orchestrator in control mode
• Manage PR lifecycle with multiple agents
• Watch PRs with automatic conflict resolution and comment handling
• Coordinate autonomous multi-agent workflows

Capabilities

Orchestrator Control

• Spawn a master control thread
• Manage orchestrator daemon lifecycle
• Monitor system health and thread status
• Handle escalations from sub-agents

PR Lifecycle Management

• Watch PRs with full lifecycle management
• Automatic merge conflict resolution
• Review comment handling (respond + resolve)
• Configurable auto-merge behavior

Multi-Agent Spawning

• PR Shepherd agents per PR
• Merge Conflict Resolver agents
• Review Comment Handler agents
• All working in isolated worktrees

Essential Commands

# Start orchestrator control thread
ct control start [--interactive] [--auto-merge]

# Check control status
ct control status

# Stop control
ct control stop

# Watch a PR with full lifecycle management
ct pr watch <number> [--auto-merge] [--interactive] [--poll-interval 30]

# Check PR lifecycle status
ct pr status <number>

# View PR comment status
ct pr comments <number>

# List all watched PRs
ct pr list

Configuration Flags

PR Completion Criteria

A PR is considered "done" when ALL are true:

• CI passing (all checks green)
• No merge conflicts
• All review comments RESPONDED
• All review comments RESOLVED (threads closed)
• Required approvals received

Proactive Review State Detection

CRITICAL: Do NOT use "pending" states that wait passively. Always check proactively:

# Check IMMEDIATELY after Copilot review arrives
UNRESOLVED=$(gh api graphql -f query='query {
  repository(owner: "hanibalsk", name: "property-management") {
    pullRequest(number: <PR_NUMBER>) {
      reviewThreads(first: 100) { nodes { isResolved } }
    }
  }
}' | jq '[.data.repository.pullRequest.reviewThreads.nodes[] | select(.isResolved == false)] | length')

if [ "$UNRESOLVED" -gt 0 ]; then
  # Spawn review-comment-handler agent IMMEDIATELY
  # Do NOT mark as "pending" and wait
fi

Auto-Approval Workflow

Use the GitHub Actions workflows for approval:

# Trigger auto-approve (waits for CI, checks conditions)
gh workflow run auto-approve.yml -f pr_number=<PR_NUMBER>

# Direct approval (with options)
gh workflow run approve-pr.yml -f pr_number=<PR_NUMBER> [-f skip_ci_check=false]

When Orchestrating PRs

1. On PR Creation: Start monitoring CI
2. On CI Complete: Check for Copilot review
3. On Copilot Review: IMMEDIATELY check unresolved threads
4. If Unresolved > 0: Spawn review-comment-handler agent
5. If Unresolved == 0: Trigger auto-approve workflow
6. On Merge Conflict: Spawn merge-conflict-resolver agent
7. After All Resolved: Trigger auto-approve workflow

Multi-Agent Architecture

Control Thread (ct control start)
    |
    +-- Git Poller (polls PR status)
    |
    +-- PR Shepherd (per watched PR)
            |
            +-- Merge Conflict Agent (on conflict)
            |
            +-- Comment Handler Agent (per comment)

Example Workflows

Watch a PR with Auto-Merge

# Start orchestrator control
ct control start --auto-merge

# Watch a specific PR
ct pr watch 123 --auto-merge

# The system will:
# 1. Poll PR status every 30 seconds
# 2. Detect and auto-resolve merge conflicts
# 3. Handle all review comments (respond + resolve)
# 4. Auto-merge when all criteria are met

Interactive Mode

# Start in interactive mode
ct control start --interactive

# You'll be asked to confirm:
# - Spawning new PR shepherds
# - Auto-merging PRs
# - Handling escalations

Monitor Multiple PRs

ct control start

ct pr watch 123
ct pr watch 124
ct pr watch 125

# Check status of all
ct pr list

# Detailed status
ct pr status 123

Event Types

Published by Control Thread

• ORCHESTRATOR_STARTED - Control thread started
• SHEPHERD_SPAWNED - New PR shepherd spawned
• SYSTEM_STATUS - Periodic status update

Published by PR Shepherd

• PR_STATE_CHANGED - State machine transition
• MERGE_CONFLICT_DETECTED - Conflict needs resolution
• REVIEW_COMMENTS_PENDING - Comments need handling
• PR_READY_FOR_MERGE - All criteria met

Published by Sub-Agents

• MERGE_CONFLICT_RESOLVED - Conflict fixed
• COMMENT_RESPONDED - Reply posted
• COMMENT_RESOLVED - Thread closed

Integration with Existing Skills

This skill extends the threads skill with:

• Higher-level orchestration
• PR lifecycle management
• Multi-agent coordination
• Conflict and comment handling

Use alongside:

• bmad-autopilot for epic development
• thread-spawner for parallel thread creation

Base + Fork Pattern

Sub-agents use memory-efficient fork worktrees:

PR Base Worktree (created when watching PR)
    │
    ├── Fork: conflict-resolver (on conflict)
    │   └── Shares git objects with base
    │
    ├── Fork: comment-handler-1 (per comment)
    │   └── Shares git objects with base
    │
    └── Fork: comment-handler-2
        └── Shares git objects with base

After each fork completes:
1. Merge fork back to base
2. Remove fork
3. Push from base (once for all)

Fork Commands

# Create fork for sub-agent
ct worktree fork 123 conflict-fix fix/conflict conflict_resolution

# After sub-agent completes
ct worktree merge-back conflict-fix
ct worktree remove-fork conflict-fix

# Push from base
cd $(ct worktree base-path 123)
git push

Distributed Deployment

Run orchestrator on central server with remote workers:

# Central server
ct api start --bind 0.0.0.0 --token $TOKEN

# Worker machines
ct remote connect central:31337 --token $TOKEN
ct spawn epic-7a --template bmad-developer.md

Error Recovery

Handle Fork Merge Conflicts

if ! ct worktree merge-back my-fork; then
  # Retry with fresh fork
  ct worktree remove-fork my-fork --force
  ct worktree base-update 123
  # Re-fork and retry
fi

Reconcile Database with Filesystem

# Check for inconsistencies
ct worktree reconcile

# Auto-fix orphans
ct worktree reconcile --fix

Documentation

• ARCHITECTURE.md - System architecture
• AGENT-COORDINATION.md - Coordination patterns
• WORKTREE-GUIDE.md - Base + fork details
• EVENT-REFERENCE.md - Event types
• MULTI-INSTANCE.md - Distributed deployment