What is plan-management?

This Claude skill provides a robust multi-layer planning and playbook management system designed for long-running agent sessions. It organizes complex tasks into a clear hierarchy of roadmaps, milestones, and phases, ensuring operational consistency through automated state tracking, strict transition rules, and integrated critic verification to prevent process skipping and state drift.

When should I use plan-management?

plan-management is useful in the following scenarios: • Complex Project Orchestration: Decomposing high-level project goals into a structured hierarchy of roadmaps and milestones for long-term execution. • Standardized Workflow Governance: Enforcing strict phase transitions (e.g., pending to designing to implementing) to ensure architectural design is never skipped before coding. • Context-Aware Session Recovery: Synchronizing agent focus with git branches and state files, allowing the AI to instantly resume work by reading the current playbook and active phase. • Quality Assurance & Verification: Implementing an evidence-based completion system where tasks are only marked 'done' after passing specific criteria and a mandatory critic review.

name	plan-management
description	Multi-layer planning and playbook management. Use when creating playbooks, transitioning phases, or managing plan hierarchy. Triggers on "plan", "playbook", "phase", "roadmap", "milestone" keywords.

Plan Management Skill

Multi-layer planning system for long-running agent sessions.

Plan Hierarchy Structure

roadmap (最上位)
  └── milestones
        └── playbooks (1 task = 1 playbook = 1 branch)
              └── phases

When to Use This Skill

New task: Create playbook before starting work
Phase transition: Update playbook status and state.md
Session start: Read roadmap → playbook → understand context
Task completion: Verify against done_criteria, call critic

Playbook Creation Flow

1. Determine task scope
2. Check existing playbooks in plan/
3. Create playbook using plan/template/playbook-format.md
4. Update state.md:
   - active_playbooks.{layer}: path/to/playbook
   - layer.{layer}.playbook: path/to/playbook
5. Commit playbook

Phase Transition Rules

状態遷移:
  pending → designing → implementing → [reviewing] → state_update → done

禁止遷移:
  - pending → implementing (設計スキップ禁止)
  - pending → done (全スキップ禁止)
  - * → done without critic (自己報酬詐欺防止)

Phase 完了条件:
  1. done_criteria の全項目に証拠がある
  2. test_method を実際に実行した
  3. critic が PASS を返した

Four-Tuple Coherence

四つ組:
  - focus.current (state.md)
  - layer.state (state.md)
  - playbook (plan/playbook-*.md)
  - branch (git)

整合性ルール:
  - playbook.branch == git current branch
  - focus.current == active playbook's layer
  - layer.state reflects playbook progress

Session Start Checklist

必須 Read:
  1. state.md → focus.current 確認
  2. roadmap (if workspace layer)
  3. playbook (if session=task)
  4. project.md (if product layer)

branch 確認:
  - main なら新ブランチ作成
  - playbook.branch と一致するか確認

playbook 確認:
  - null なら /playbook-init 実行
  - 存在するなら Read して in_progress phase 特定

Automatic Triggers

This skill activates when Claude detects:

"計画を立てて" / "plan" / "playbook"
"次のフェーズ" / "phase"
"ロードマップ" / "roadmap" / "milestone"
Session start with session=task

Integration with Hooks

session-start.sh:
  - Outputs required Read list
  - Warns if playbook=null

check-coherence.sh:
  - Validates four-tuple alignment
  - Blocks commits if misaligned

session-end.sh:
  - Updates session_tracking
  - Reminds about uncommitted changes

Best Practices

One task = One playbook = One branch
Read before write: Always read playbook before modifying
Evidence-based completion: No done without proof
Critic before done: Always call critic agent
Commit after each phase: Keep git in sync

plan-management

When & Why to Use This Skill

Use Cases