Understand the Whole:
- Define project goal in 1-2 sentences
- Identify the boundary: what's in, what's explicitly out
- Understand current state (existing code, infrastructure, constraints)
Identify Sub-Tasks:
- Break the project into the minimum number of independent work units
- Each sub-task should be completable within a single workflow cycle (analyze → plan → execute)
- Name each sub-task (2-4 words, English, hyphen-separated)
Analyze Dependencies:
- Map which sub-tasks block others
- Identify what can run in parallel
- Find the critical path
Assess Risks:
- Technical uncertainties requiring validation (/workflow-validate)
- External dependencies or unknowns
- Areas where the decomposition itself is uncertain
ADR Assessment:
- After decomposition, evaluate whether any project-level decisions meet ADR criteria (PRICE)
- Epic-level decisions often qualify: technology choices, architectural patterns, cross-cutting concerns
- If any PRICE criterion applies, add the ADR assessment section to the document
Assess Decomposition Confidence:
- Apply Decision Honesty rule (same as workflow-analyze)
- Confident: Clear boundaries, well-understood domain
- Tentative: Some boundaries may shift after early tasks reveal more
Write Document:
- Create docs/epic/{project-name}/analysis.md (Korean)

Key Rules

Documentation Language

CRITICAL: All documents must be written in Korean.

Thinking Order

Understand the whole project first. Decompose second. Premature decomposition without understanding the whole leads to wrong boundaries.

Decomposition Principles (CRITICAL)

Minimum viable split: Fewest sub-tasks that cover the full scope. If two tasks always change together, they're one task.
Session-independent: Each sub-task must be executable in a separate Claude Code session with only its own context (analysis.md, plan.md) — no implicit dependency on "what the previous session knew."
Vertical, not horizontal: Don't split by layer (DB → API → UI). Split by capability or domain boundary.
Size calibration:
- Each sub-task = 1-5 commits when planned with /workflow-plan
- If a sub-task feels like 1 commit → it's too small, merge with neighbors
- If a sub-task feels like 10+ commits → it needs further splitting
Workflow-ready output: Each sub-task description must be sufficient as input to /workflow-analyze {task-name}

Anti-Bias Rules (CRITICAL)

Core: Don't over-engineer the decomposition itself. The goal is clarity, not perfection.

Over-decomposition bias
- AI tends to split too much ("just to be safe")
- Ask: "Would merging these two tasks lose anything meaningful?"
- If the answer is no → merge them
Scope creep bias
- AI tends to add "nice to have" sub-tasks not in the original request
- Every sub-task must trace back to the user's stated goal
- If it doesn't → exclude it or mark it explicitly as "future consideration"
Perfectionism bias
- AI tends to add "setup" and "cleanup" tasks that aren't needed
- Infrastructure/tooling tasks are only valid if they unblock other tasks
- Don't create sub-tasks for things the team already knows how to do
Symmetry bias
- AI tends to make all sub-tasks similar in size
- Uneven size distribution is natural — don't force uniformity
Decision Honesty (same principle as workflow-analyze)
- Confident: Domain is well-understood, boundaries are clear → state definitively
- Tentative: Some boundaries may shift → mark uncertain areas, recommend revisiting after Phase 1
- Default is Tentative. Confident requires well-understood domain + clear boundaries.

Must Do

Project goal in 1-2 sentences
Minimum number of sub-tasks (prefer fewer, larger tasks)
Clear dependency graph
Each sub-task named and described enough for /workflow-analyze input
Estimate relative size (S/M/L) per sub-task
Identify what needs /workflow-validate before proceeding
Mark uncertain boundaries explicitly

Must Not Do

Create sub-tasks for obvious setup (project init, linter config, etc.) unless genuinely complex
Split by technical layer (types → logic → tests → UI)
Add tasks not in the original scope without explicit marking
Estimate time in hours/days (use relative size only)
Over-specify implementation details (that's workflow-analyze's job)

Document Template

File to create: docs/epic/{project-name}/analysis.md (Korean)

# [Project Name] - Epic Analysis

## Overview

**Project Goal**: [1-2 sentences — what this project achieves when complete]
**Trigger**: [Why now? What prompted this project?]

---

## Scope

**Included**:

- [What's in scope]

**Excluded**:

- [What's explicitly out — and why]

**Current State**: [Brief description of what exists today]

---

## Sub-Task Decomposition

<!-- Confident: use this format -->

**Confidence**: Confident (boundaries are clear)

<!-- Tentative: use this format instead -->

**Confidence**: Tentative (boundaries may shift)
**Uncertain Areas**: [Which sub-task boundaries might change and why]
**Revisit After**: [Which task(s) should complete before re-evaluating]

### Task 1: [name] (Size: S/M/L)

**Goal**: [1 sentence — what this task achieves independently]
**Why separate**: [Why this isn't part of another task]
**Workflow input**: [Key context for `/workflow-analyze {name}` — 2-3 sentences max]

### Task 2: [name] (Size: S/M/L)

[Same structure]

### Task N: [name] (Size: S/M/L)

[Same structure]

---

## Dependencies

Task 1 ──→ Task 3 ──→ Task 5 Task 2 ──→ Task 4 ──┘


**Critical Path**: Task 1 → Task 3 → Task 5
**Parallelizable**: Task 1 and Task 2 can proceed simultaneously

---

## Risks & Validation Needs

**Needs `/workflow-validate`**:

- [Technical uncertainty 1]: [What to validate and why]

**External Dependencies**:

- [Dependency]: [Impact if unavailable]

---

## Suggested Execution Order

**Phase 1**: [Task names — foundation]
**Phase 2**: [Task names — core features]
**Phase 3**: [Task names — integration/polish]

> Start with: `/workflow-analyze {task-1-name}`

---

## ADR Assessment

<!-- Include this section ONLY when project-level decisions meet PRICE criteria -->
<!-- Omit entirely if no criterion applies -->

**Decisions requiring ADR**:

- [Decision 1]: PRICE criteria [R, I] — [brief reason]
- [Decision 2]: PRICE criteria [P] — [brief reason]

**Recommendation**: Create ADRs before starting sub-task execution
**Suggested Timing**: After epic analysis approval, before first `/workflow-analyze`

> To create: `/adr [topic]`

---

## Self-Check

- [ ] Minimum number of sub-tasks (couldn't merge any two without losing independence)
- [ ] Every sub-task traces to the stated project goal
- [ ] No "nice to have" tasks smuggled in
- [ ] Each sub-task is session-independent (no implicit context dependency)
- [ ] Vertical split, not horizontal (each task delivers a complete capability)
- [ ] Decomposition confidence honestly assessed

Execution

Now start the task according to the guidelines above.

epic-analyze

Epic Analysis Command

User Input

Outline