eoa-orchestration-commands

Orchestration Commands Skill

Overview

This skill teaches how to use the orchestration phase commands in the Orchestrator Agent plugin. These commands manage the execution loop that coordinates remote agents to implement approved plans.

Instructions

1. Start an orchestration phase to implement an approved plan
2. Monitor the status of modules, agents, and assignments
3. Check the orchestrator loop state and pending tasks
4. Control or cancel the orchestrator loop
5. Understand how the stop hook prevents premature exit

Output

Output Type	Format	Example
Status report	Markdown table	Module completion percentages, agent assignments
Loop state	Text summary	Active/inactive, iteration count, pending tasks
Cancellation	Confirmation	"Orchestrator loop cancelled at iteration X"
Error messages	Text	Hook blocking reasons, state file issues

Prerequisites

Before using orchestration commands, ensure:

1. Plan Phase is complete (via /approve-plan)
2. The state file design/state/exec-phase.md exists
3. You understand the Two-Phase Mode workflow (Plan Phase then Orchestration Phase)

---

Command Summary

Command	Purpose	When to Use
/start-orchestration	Activate orchestration phase	After plan approval, to begin implementation
/orchestration-status	View module/agent progress	To monitor implementation progress
/orchestrator-status	Check loop state and tasks	To see pending tasks across all sources
/orchestrator-loop	Start continuous task loop	To activate task-driven development loop
/cancel-orchestrator	Cancel active loop	To manually stop the orchestrator loop

Examples

Example 1: Start Orchestration After Plan Approval

# Step 1: Start orchestration with GitHub Project sync
/start-orchestration --project-id PVT_kwDOB1234567

# Step 2: Register your agents
/register-agent ai implementer-1 --session helper-agent-generic

# Step 3: Assign first module
/assign-module auth-core implementer-1

# Step 4: Monitor progress
/orchestration-status

Example 2: Start Orchestrator Loop

# Start continuous task-driven development
/orchestrator-loop "Complete all pending authentication tasks" --max-iterations 50

# Check loop status
/orchestrator-status --verbose

# Cancel if needed
/cancel-orchestrator

---

Quick Reference: Starting Orchestration

If you need to start the orchestration phase after plan approval:

1. Run /start-orchestration (optionally with --project-id for GitHub sync)
2. Register agents with /register-agent
3. Assign modules with /assign-module
4. Monitor with /orchestration-status every 10-15 minutes

See references/start-orchestration-procedure.md:

• 1.1 When to start orchestration phase
• 1.2 Prerequisites verification checklist
• 1.3 Command syntax and options
• 1.4 Post-start agent registration workflow
• 1.5 GitHub Project integration setup

---

Quick Reference: Monitoring Status

If you need to check implementation progress:

Use /orchestration-status to see:

• Module completion percentages
• Registered agents and their assignments
• Instruction verification status
• Progress polling history

See references/status-monitoring.md:

• 2.1 Understanding the orchestration status output
• 2.2 Reading module status indicators
• 2.3 Interpreting agent registry information
• 2.4 Tracking active assignments and polling
• 2.5 Using verbose mode for detailed diagnostics

---

Quick Reference: Orchestrator Loop Control

If you need to start or manage the orchestrator loop:

The orchestrator loop monitors multiple task sources and prevents exit until ALL are complete:

• Claude Tasks via TaskList API
• GitHub Projects open items
• Task file markdown checklists
• Claude TODO list (session)

See references/orchestration-loop-mechanics.md:

• 3.1 What the orchestrator loop does
• 3.2 Task source monitoring and priority
• 3.3 Iteration counting and max iterations
• 3.4 Verification mode (4-loop quadruple-check)
• 3.5 Completion signals (ALL_TASKS_COMPLETE)
• 3.6 Stop hook behavior and blocking logic

---

Quick Reference: Cancellation

If you need to cancel the orchestrator loop:

Use /cancel-orchestrator to:

1. Check if a loop is active
2. Read current iteration from state file
3. Remove the state file to stop the loop

See references/cancellation-cleanup.md:

• 4.1 When to cancel vs let loop complete naturally
• 4.2 Cancellation procedure step-by-step
• 4.3 Cleanup of state files and locks
• 4.4 Recovery after unexpected termination

---

State File Reference

If you need to understand or debug state files:

The orchestrator uses markdown files with YAML frontmatter to track state:

• design/state/loop.md - Loop state
• design/state/exec-phase.md - Execution phase state

See references/state-file-format.md:

• 5.1 Loop state file format and fields
• 5.2 Execution phase state file format
• 5.3 Frontmatter field definitions
• 5.4 How to manually edit state files
• 5.5 State file corruption recovery

---

Troubleshooting

Error Handling

See references/troubleshooting.md:

• 6.1 Loop won't start - common causes
• 6.2 Stop hook not firing - debugging steps
• 6.3 Tasks showing as pending incorrectly
• 6.4 Lock file issues and stale locks
• 6.5 Concurrent execution conflicts
• 6.6 Verification mode stuck
• 6.7 Helper script failures

---

Python Scripts Reference

The following scripts implement the orchestration commands. Located in the plugin's scripts/ directory:

Script	Purpose	Used By
eoa_start_orchestration.py	Activates orchestration phase	/start-orchestration
eoa_orchestration_status.py	Displays phase status	/orchestration-status
eoa_check_orchestrator_status.py	Shows loop state	/orchestrator-status
eoa_setup_orchestrator_loop.py	Creates loop state file	/orchestrator-loop
eoa_orchestrator_stop_check.py	Stop hook enforcement	Hook event

---

Command Details

/start-orchestration

Purpose: Enter Orchestration Phase to coordinate remote agents implementing the approved plan.

Syntax:

/start-orchestration [--project-id PVT_kwDOBxxxxxx]

Options:

• --project-id: GitHub Project ID for Kanban synchronization

What it does:

1. Verifies Plan Phase is complete
2. Sets orchestration status to "executing"
3. Loads module list from approved plan
4. Enables stop hook enforcement
5. Prepares agent tracking structures

Example:

/start-orchestration --project-id PVT_kwDOB1234567

---

/orchestration-status

Purpose: View current Orchestration Phase progress including modules, agents, and verification status.

Syntax:

/orchestration-status [--verbose] [--agents-only] [--modules-only]

Options:

• --verbose: Show detailed polling history and acceptance criteria
• --agents-only: Show only agent information
• --modules-only: Show only module status

Output sections:

• Phase status header (plan ID, status, progress percentage)
• Module status table with completion indicators
• Registered agents list (AI and human)
• Active assignments with verification and polling info

---

/orchestrator-status

Purpose: Check orchestrator loop status and pending tasks across all sources.

Syntax:

/orchestrator-status [--verbose]

Options:

• --verbose: Show detailed debug information and recent log entries

Output sections:

• Loop active/inactive status
• Iteration count and limits
• Task sources with pending counts (Claude Tasks, GitHub, Task file, TODO)
• Current task preview
• Debug info (log file, lock file status)

---

/orchestrator-loop

Purpose: Start orchestrator loop for continuous task-driven development.

Syntax:

/orchestrator-loop [PROMPT] [options]

Options:

• --max-iterations N: Maximum iterations before escalation (default: 100)
• --completion-promise TEXT: Promise phrase to trigger completion
• --task-file PATH: Markdown task file to monitor
• --check-tasks BOOL: Check Claude Tasks (default: true)
• --check-github BOOL: Check GitHub Projects (default: true)
• --github-project ID: Specific GitHub Project ID

Example:

/orchestrator-loop "Complete all pending authentication tasks" --max-iterations 50

---

/cancel-orchestrator

Purpose: Cancel active orchestrator loop.

What it does:

1. Checks if loop state file exists
2. Reads current iteration number
3. Removes state file to stop loop
4. Reports cancellation with iteration count

Note: This command is hidden from slash command suggestions but remains functional.

---

Stop Hook Behavior

The orchestrator stop hook (eoa_orchestrator_stop_check.py ) enforces completion requirements:

Blocking conditions:

• Plan Phase incomplete (requirements not documented, plan not approved)
• Orchestration Phase incomplete (modules not implemented)
• Pending tasks in any monitored source
• Instruction verification incomplete
• Config feedback requests unresolved
• Verification loops remaining (4 required after all tasks complete)

Completion signals:

• Output ALL_TASKS_COMPLETE when all tasks genuinely done
• Output <promise>YOUR_PHRASE</promise> matching configured promise

Recovery behavior:

• Fail-safe exit on unrecoverable errors (prevents user trap)
• Conservative blocking when task status cannot be determined
• Retry logic for transient failures
• Lock file cleanup for stale processes

---

Examples

Example 1: Complete Orchestration Start

# Step 1: Start orchestration with GitHub Project sync
/start-orchestration --project-id PVT_kwDOB1234567

# Step 2: Register AI agent
/register-agent ai implementer-1 --session helper-agent-generic

# Step 3: Assign first module
/assign-module auth-core implementer-1

# Step 4: Monitor progress
/orchestration-status --verbose

Example 2: Check Orchestrator Loop Status

# Start loop
/orchestrator-loop "Complete all authentication tasks" --max-iterations 50

# Check status
/orchestrator-status --verbose

# If needed, cancel
/cancel-orchestrator

Example 3: Monitoring During Implementation

# Check all agents every 10-15 minutes
/check-agents

# Or check specific agent
/check-agents --agent implementer-1

# View orchestration status
/orchestration-status --modules-only

---

Checklist: Starting Orchestration

Copy this checklist and track your progress:

• Plan Phase complete (/approve-plan executed)
• State file design/state/exec-phase.md exists
• Run /start-orchestration with optional --project-id
• Register AI agents with /register-agent ai <agent_id> --session <session>
• Assign modules with /assign-module <module_id> <agent_id>
• Execute Instruction Verification Protocol for each agent
• Begin polling with /check-agents every 10-15 minutes

---

Checklist: Monitoring Progress

Copy this checklist and track your progress:

• Run /orchestration-status to see module completion
• Check for agents with incomplete instruction verification
• Review polling history for stuck agents
• Use --verbose flag for detailed diagnostics
• Check /orchestrator-status for pending tasks across sources

---

Checklist: Cancellation

Copy this checklist and track your progress:

• Confirm you want to cancel (tasks may be incomplete)
• Run /cancel-orchestrator
• Verify state file removed
• Check no orphaned lock files in .claude/
• If needed, manually remove design/state/loop.md

---

Resources

• start-orchestration-procedure.md - Starting orchestration
• status-monitoring.md - Reading status output
• orchestration-loop-mechanics.md - Loop behavior
• cancellation-cleanup.md - Cancellation procedures
• state-file-format.md - State file schemas
• troubleshooting.md - Common issues and solutions