eoa-two-phase-mode

name	eoa-two-phase-mode
description	Use when orchestrating complex projects. Trigger with multi-module or two-phase planning requests.
license	Apache-2.0
compatibility	Requires AI Maestro messaging system, GitHub CLI (gh), remote agents registered by user, and YAML frontmatter state files. Requires AI Maestro installed.
metadata	{"author":"Emasoft","version":"2.6.0"}
context	fork
user-invocable	false
agent	eoa-main
workflow-instruction	Step 10
procedure	proc-decompose-design

Two-Phase Mode Skill

Overview

Two-Phase Mode separates orchestration into two distinct phases:

Prerequisites

• AI Maestro messaging system running
• GitHub CLI (gh) authenticated
• Remote agents registered by user
• Understanding of YAML frontmatter state files

Output

Output Type	Location	Description
Plan Phase State	.claude/orchestrator-plan-phase.local.md	YAML frontmatter with requirements, modules, criteria
Orchestration State	.claude/orchestrator-orchestration-phase.local.md	YAML frontmatter with agent assignments, module status
Design Documents	design/<platform>/	Templates, handoffs, RDD files, specs
GitHub Issues	GitHub repository	Module issues created from approved plan
Claude Tasks	Claude Code session	Persistent task tracking across compacting

Instructions

1. Understand the two phases: Plan Phase (requirements) and Orchestration Phase (implementation).

   Phase	Purpose	Activities	Exit Condition
   Plan Phase	Write requirements	Design specs, architecture, task breakdown	All requirements documented + user approval
   Orchestration Phase	Direct implementation	Coordinate remote agents module by module	All modules implemented + 4 verification loops

2. Start Plan Phase: Use /start-planning command to create state file and begin requirements documentation.

3. Document requirements: Write USER_REQUIREMENTS.md, define architecture, break down modules, set acceptance criteria.

4. Approve plan: Use /approve-plan to transition to Orchestration Phase and create GitHub issues.

5. Start Orchestration Phase: Use /start-orchestration to initialize orchestration state file.

6. Register agents: Use /register-agent to register remote AI agents or human developers.

7. Assign modules: Use /assign-module to assign modules to registered agents.

8. Verify instructions: Execute Instruction Verification Protocol before authorizing implementation.

9. Poll progress: Use /check-agents every 10-15 minutes to actively poll implementers.

10. Handle updates: If requirements change, use Instruction Update Verification Protocol.

11. Complete verification loops: Each module requires 4 verification loops before completion.

12. Exit when complete: Stop hook ensures all modules are complete before allowing exit.

Core Principle: Dynamic Flexibility with Completion Enforcement

The orchestrator enforces completion of CURRENT tasks while allowing dynamic modifications.

Aspect	Behavior
User can always	Add features, change requirements, remove pending items
System enforces	All CURRENT tasks must complete before stopping
Dynamic tracking	Stop hook checks CURRENT state, not original state

How it works: If user adds OAuth2 module, stop hook blocks until OAuth2 is ALSO complete. If user removes OAuth2 (before started), it's no longer tracked.

---

Core Principle: Claude Tasks Scheduling (CRITICAL)

MANDATORY: All instructions must be immediately scheduled as Claude Tasks. This is non-negotiable.

Principle	Description
Immediate scheduling	When receiving any instruction, create Claude Tasks FIRST before acting
Persistence across compacting	Claude Tasks persist via TaskList API, surviving context compacting
Series closure	Every task series ends with: verification checklist, archive record, commit
Recovery capability	Agents can recover state after compacting by reading Claude Tasks

Why this matters: Claude Code context windows compact during long sessions. Without Claude Tasks, agents forget instructions and progress. Claude Code native Tasks provide persistent memory.

See references/native-task-persistence.md for complete documentation.

---

Reference Files

Plan Phase Workflow (references/plan-phase-workflow.md)

Complete workflow for Plan Phase Mode.

Section	Topics
1. Entering Plan Phase	/start-planning command, state file initialization
2. Planning Activities	User goals, USER_REQUIREMENTS.md, architecture, modules, acceptance criteria
3. Modifying the Plan	/add-requirement, /modify-requirement, /remove-requirement
4. Plan Phase Completion	Exit criteria checklist, /approve-plan, GitHub Issues creation

When to use: Starting a new project, entering planning mode, modifying requirements.

---

Orchestration Phase Workflow (references/orchestration-phase-workflow.md)

Complete workflow for Orchestration Phase Mode.

Section	Topics
1. Entering Orchestration Phase	/start-orchestration command, state file structure
2. Agent Registration	/register-agent, AI agents vs human developers, agent types
3. Module Assignment	/assign-module, /reassign-module
4. Monitoring Progress	/orchestration-status, /check-agents for polling
5. Modifying During Orchestration	/add-module, /modify-module, /remove-module, /prioritize-module
6. Completion and Exit	All modules complete criteria, 4-verification loops, stop hook

When to use: After plan approval, during module implementation, managing remote agents.

Prerequisite: Plan Phase Workflow

---

Instruction Verification Protocol (references/instruction-verification-protocol.md)

MANDATORY protocol before ANY remote agent begins implementation.

Section	Topics
1. Why This Protocol Exists	Reasons for misinterpretation, orchestrator responsibility
2. The 8-Step Protocol Flow	Send assignment, request repetition, verify, answer questions, authorize
3. Message Templates	Initial assignment, correction, question resolution, authorization
4. Tracking Verification Status	State file fields, status values
5. Failure Conditions	When NOT to authorize, action on failure

When to use: After assigning a module, before authorizing implementation.

Prerequisite: Orchestration Phase Workflow

---

Proactive Progress Polling Protocol (references/proactive-progress-polling.md)

MANDATORY polling protocol with 6 required questions every 10-15 minutes.

Section	Topics
1. Why This Protocol Exists	Never assume "no news is good news", orchestrator MUST actively ask
2. The 6 Mandatory Poll Questions	Progress, next steps, issues, unclear items, difficulties, needs
3. Poll Message Template	Full template with all 6 questions
4. Response Actions	Action table by response type, Adapt-or-Escalate decision tree
5. Poll Tracking	State file polling fields, poll history structure

When to use: Every 10-15 minutes during active implementation, when agent silent too long.

Prerequisite: Instruction Verification Protocol

---

Instruction Update Verification Protocol (references/instruction-update-verification-protocol.md)

MANDATORY protocol whenever sending UPDATED instructions to an implementer who is ALREADY working.

Section	Topics
1. When This Protocol Applies	Triggers (requirement change, design update, spec clarification)
2. The 5-Step Update Verification Flow	Send update, confirm receipt, assess feasibility, address concerns, authorize resume
3. Message Templates	Update notification, feasibility request, concern resolution, resume auth
4. Tracking Update Verification	State file fields, status values
5. Special Cases	Minor clarifications, major design changes, user requirement changes
6. Configuration Feedback Loop	When implementer needs config changes from orchestrator

When to use: When sending any update mid-implementation, changing requirements, handling config feedback.

Prerequisite: Proactive Progress Polling Protocol

---

State File Formats (references/state-file-formats.md)

Complete YAML frontmatter specifications for both phase state files.

Section	Topics
1. Plan Phase State File	Location, complete YAML schema, field descriptions
2. Orchestration Phase State File	Location, complete YAML schema, field descriptions
3. Agent Assignment Structure	Assignment fields, verification tracking, polling tracking
4. Module Status Structure	Module fields, status values

When to use: Creating/parsing state files, understanding structure, debugging state issues.

---

Design Folder Structure (references/design-folder-structure.md)

Standardized folder structure for design documents, templates, handoffs, and RDD files.

Section	Topics
1. Why a Standardized Structure	Git tracking, per-platform organization, single source of truth
2. Folder Structure Specification	Root location (design/), per-platform structure, directory tree
3. File Types and Locations	Templates, handoffs, RDD files, config files, specs
4. Usage Workflow	Creating files, compiling templates, storing responses
5. Git Tracking Rules	What to track, what to gitignore
6. Multi-Platform Projects	Shared resources, platform-specific customization

When to use: Setting up design folder, creating templates, compiling handoffs.

Prerequisite: Plan Phase Workflow

---

Command Reference (references/command-reference.md)

Complete reference for all 16 Two-Phase Mode commands.

Section	Commands
1. Plan Phase Commands (6)	/start-planning, /planning-status, /add-requirement, /modify-requirement, /remove-requirement, /approve-plan
2. Orchestration Phase Commands (10)	/start-orchestration, /orchestration-status, /register-agent, /assign-module, /add-module, /modify-module, /remove-module, /prioritize-module, /reassign-module, /check-agents

When to use: Quick reference for command syntax and parameters.

---

Script Reference (references/script-reference.md)

Complete reference for all Two-Phase Mode Python scripts.

Section	Scripts
1. Plan Phase Scripts (4)	eoa_start_planning.py, eoa_planning_status.py, eoa_modify_requirement.py, eoa_approve_plan.py
2. Orchestration Phase Scripts (14)	eoa_start_orchestration.py, eoa_orchestration_status.py, eoa_register_agent.py, eoa_assign_module.py, eoa_modify_module.py, eoa_reassign_module.py, eoa_check_remote_agents.py, eoa_notify_agent.py, eoa_check_plan_phase.py, eoa_check_orchestration_phase.py, eoa_sync_github_issues.py, eoa_verify_instructions.py, eoa_poll_agent.py, eoa_update_verification.py
3. Design & GitHub Scripts (5)	eoa_init_design_folders.py, eoa_compile_handoff.py, eoa_design_search.py, eoa_sync_kanban.py, eoa_create_module_issues.py
4. Modified Scripts (1)	eoa_orchestrator_stop_check.py (phase-aware)

When to use: Understanding script functionality, debugging, learning parameters.

---

Claude Tasks Scheduling Principle (references/native-task-persistence.md)

CRITICAL principle for all orchestrator and subagent operations.

Section	Topics
1. Why This Principle Exists	Context compacting problem, how Claude Tasks solve it
2. The Claude Tasks Scheduling Rule	When to create Claude Tasks, what to include
3. Task Series Structure	Action tasks, verification, archive, commit, series closure
4. Commit Message Format	[SERIES-COMPLETE] format, required sections
5. Subagent Claude Tasks Requirements	Orchestrator/subagent responsibilities, Claude Tasks location
6. Integration with Stop Hook	How stop hook checks Claude Tasks, block conditions

When to use: When receiving ANY instruction, spawning subagents, completing task series, recovering after compacting.

Prerequisite: None (CRITICAL - read first)

---

Issue Handling Workflow (references/issue-handling-workflow.md)

Complete workflow for handling implementer-reported issues.

Section	Topics
1. When to Trigger	AI Maestro messages, progress polls, code review, test failures
2. Issue Categories	BUG, BLOCKER, QUESTION, ENHANCEMENT, CONFIG, INVESTIGATION
3. Creating Issue Tasks	/create-issue-tasks command, eoa_create_issue_tasks.py
4. Standard Task Workflow	Assessment, triage, investigation, test creation, GitHub workflow, resolution
5. Category-Specific Workflows	BUG, BLOCKER, QUESTION, ENHANCEMENT, CONFIG, INVESTIGATION workflows
6. Integration with Stop Hook	Issue task file checking, open issue blocking

When to use: When implementer reports issue, poll reveals problem, code review finds bug, tests fail.

Prerequisite: Claude Tasks Scheduling Principle

---

Workflow Diagram (references/workflow-diagram.md)

Visual representation of the complete Two-Phase Mode workflow.

Section	Topics
1. Visual Workflow Overview	Complete ASCII flowchart from user goal to exit
2. Phase Transitions	Transition triggers and conditions table
3. Module Processing Loop	Per-module flowchart with all protocols

When to use: Understanding the overall flow, visualizing phase transitions, seeing module loop.

---

Quick Reference Checklist (references/quick-reference-checklist.md)

Actionable checklists for all phases and operations.

Section	Topics
1. Plan Phase Checklist	All steps from /start-planning to /approve-plan
2. Orchestration Phase Checklist	All steps from /start-orchestration to exit
3. Module Completion Checklist	Assignment, verification, polling, updates, completion
4. Exit Verification Checklist	Plan Phase and Orchestration Phase exit criteria

When to use: As a quick reference while working, ensuring all steps completed.

---

Troubleshooting (references/troubleshooting.md)

Solutions for common issues in Two-Phase Mode.

Section	Topics
1. Plan Phase Issues	Won't transition, validation fails, duplicate issues
2. Orchestration Phase Issues	Stop hook blocks, module stuck, verification loops
3. State File Issues	Corruption, YAML parse error, sync issues
4. Communication Issues	Agent not responding, messages not delivered, notifications
5. Stop Hook Issues	Not firing, allows exit incorrectly, blocks incorrectly

When to use: When encountering errors, debugging issues, resolving problems.

---

Agent Types

Agent Type	Location	Assignment Method	Communication
Local Subagents	Same Claude Code	Spawned by orchestrator	Direct (Task tool)
Remote AI Agents	Independent sessions	User-provided agent IDs	AI Maestro messages
Human Developers	GitHub	GitHub Project Kanban	GitHub notifications

Important: Only user-provided agent IDs should be involved, NOT all agents on AI Maestro network.

---

Stop Hook Enforcement

The stop hook (eoa_orchestrator_stop_check.py) is phase-aware and enforces:

Phase	Blocks Exit If
Plan Phase	plan_phase_complete: false
Orchestration Phase	Any module not complete
Orchestration Phase	Verification loops remaining > 0

Dynamic enforcement: Stop hook always checks CURRENT state. If user adds/removes modules, the new state is what gets enforced.

---

File Structure

two-phase-mode/
├── SKILL.md                                    (this file)
├── README.md                                   (skill overview)
└── references/
    ├── plan-phase-workflow.md                  (Plan Phase details)
    ├── orchestration-phase-workflow.md         (Orchestration Phase details)
    ├── instruction-verification-protocol.md    (8-step initial verification)
    ├── instruction-update-verification-protocol.md  (5-step mid-impl updates)
    ├── proactive-progress-polling.md           (6 mandatory questions)
    ├── design-folder-structure.md              (design/ folder organization)
    ├── state-file-formats.md                   (YAML schemas)
    ├── command-reference.md                    (all 16 commands)
    ├── script-reference.md                     (all 16+ scripts)
    ├── native-task-persistence.md              (CRITICAL - task persistence via Claude Tasks)
    ├── issue-handling-workflow.md              (issue categories and workflows)
    ├── workflow-diagram.md                     (visual flowcharts)
    ├── quick-reference-checklist.md            (actionable checklists)
    └── troubleshooting.md                      (common issues and solutions)

---

Next Steps

1. Start with Plan Phase Workflow to enter planning mode
2. After plan approval, read Orchestration Phase Workflow
3. MANDATORY: Read Instruction Verification Protocol before assigning modules
4. MANDATORY: Read Proactive Progress Polling for monitoring
5. MANDATORY: Read Instruction Update Verification Protocol before sending mid-implementation updates
6. Reference State File Formats for state structure
7. Use Command Reference for quick command lookup
8. Use Script Reference for script details
9. See Workflow Diagram for visual overview
10. Keep Quick Reference Checklist handy during operations
11. Consult Troubleshooting when issues arise

---

Examples

Example 1: Complete Two-Phase Workflow

# PLAN PHASE
/start-planning
# Document requirements in USER_REQUIREMENTS.md
# Define architecture and modules
/approve-plan

# ORCHESTRATION PHASE
/start-orchestration --project-id PVT_kwDOB1234567
/register-agent ai implementer-1 --session helper-agent-generic
/assign-module auth-core implementer-1
# Execute Instruction Verification Protocol
# Poll every 10-15 minutes with /check-agents
# After 4 verification loops, approve PR

Example 2: Dynamic Module Addition

# User requests new feature during orchestration
/add-module "OAuth2 Support" --criteria "Google and GitHub OAuth" --priority high

# Stop hook now blocks until OAuth2 is ALSO complete
# Assign to available agent
/assign-module oauth2-support implementer-2

---

Error Handling

See Troubleshooting for complete solutions.

Issue	Cause	Resolution
Plan Phase won't transition	Requirements incomplete	Complete all required fields
Stop hook blocks exit	Modules incomplete	Finish all pending modules
Agent not responding	Connection issue	Check AI Maestro health
State file corrupt	YAML syntax error	Restore from git or recreate

---

Checklist

Copy this checklist and track your progress:

Plan Phase

• Run /start-planning command
• Document user requirements in USER_REQUIREMENTS.md
• Define architecture and design
• Break down into modules with acceptance criteria
• Review plan with user
• Run /approve-plan to transition
• Verify GitHub issues created

Orchestration Phase

• Run /start-orchestration command
• Register all remote agents with /register-agent
• Assign modules to agents with /assign-module
• Execute Instruction Verification Protocol for each module
• Poll agents every 10-15 minutes with /check-agents
• Handle any instruction updates with Update Verification Protocol
• Complete 4 verification loops per module
• Verify all modules marked complete
• Confirm stop hook allows exit

Claude Tasks Scheduling

• Create Claude Tasks immediately when receiving instructions
• Include verification task in each series
• Include archive task in each series
• Include commit task with [SERIES-COMPLETE] format
• Verify Claude Tasks persist across compacting

---

Resources

• Workflow Diagram
• Plan Phase Workflow
• Orchestration Phase Workflow
• Instruction Verification Protocol
• Instruction Update Verification Protocol
• Proactive Progress Polling
• Native Task Persistence
• Issue Handling Workflow
• Command Reference
• Script Reference
• State File Formats
• Design Folder Structure
• Quick Reference Checklist
• Troubleshooting