Outline

Parse User Input:
- Extract project name from first word of $ARGUMENTS
- Extract additional requirements from remaining text
- If project name missing, ERROR: "Please provide project name: /epic-plan PROJECT-NAME"
Check Prerequisites:
- Verify docs/epic/{project-name}/analysis.md exists
- If not exists, ERROR: "Run /epic-analyze first"
Load Analysis:
- Extract sub-task list, dependencies, and risks from analysis.md
- Note user's feedback or adjustments from conversation
Finalize Execution Order:
- Confirm or adjust phase structure from analysis.md
- Incorporate user feedback and additional requirements
- Define concrete milestones with check conditions
Prepare Task Execution Guides:
- For each sub-task, write sufficient context for /workflow-analyze in a separate session
- Include: what to read, what the prior task produced, key constraints
- This is the core value — bridging context between independent sessions
Write Document:
- Create docs/epic/{project-name}/plan.md (Korean)

Key Rules

Documentation Language

CRITICAL: All documents must be written in Korean.

Core Purpose

This document bridges context between independent sessions. Each sub-task runs in a separate Claude Code session. The plan must provide enough context for each session to start without prior knowledge.

Must Do

Reference analysis.md only (no repetition of scope/goals)
Execution guide per task: Sufficient for a cold-start session
Concrete milestones with check conditions
Specify what each task should read before starting
Specify what each task produces for downstream tasks
Progress tracking checklist

Must Not Do

Redefine sub-tasks (they're in analysis.md)
Repeat dependency graph verbatim (reference it)
Over-specify implementation (that's workflow-analyze's job)
Add new sub-tasks not in analysis.md (go back to epic-analyze if needed)
Estimate time in hours/days

Execution Guide Principles (CRITICAL)

Each task's execution guide must answer:

What to read first: Which files/docs to load for context
What prior tasks produced: Outputs from dependencies that this task needs
Key constraints: Things that must be respected (from analysis.md or user feedback)
How to start: The exact /workflow-analyze invocation with context
What this task produces: Outputs that downstream tasks will need

Parallel Execution (CRITICAL)

Default is sequential. Only mark tasks as parallel when ALL conditions are met:

No shared files: Tasks modify completely different files/modules
No git conflict risk: Merging both tasks' branches won't produce conflicts
No logical dependency: Neither task's design decisions affect the other
Truly independent domains: Different parts of the codebase with no overlap

If in doubt → sequential. The cost of unnecessary sequencing is small (slightly slower). The cost of bad parallelism is high (merge conflicts, rework, inconsistency).

Milestone Design

Define milestones at natural check points, not at every task boundary
A milestone = "after these tasks complete, verify this condition"
Keep milestones few — typically 2-4 for an entire epic
Each milestone should answer: "Are we still on track, or do we need to re-plan?"

Document Template

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

# [Project Name] - Execution Plan

> **Analysis**: See `analysis.md`

---

## Execution Order

**Dependencies**: See `analysis.md` > Dependencies section

### Phase 1: [Name]

**Tasks**: [Task names from analysis.md]
**Parallel**: [Yes — all tasks in this phase are independent / No — sequential within phase]
**Goal**: [What's true when this phase completes]

### Phase 2: [Name]

**Tasks**: [Task names]
**Parallel**: [Yes / No / Partial — Task A and B parallel, then Task C]
**Goal**: [What's true when this phase completes]

### Phase 3: [Name]

**Tasks**: [Task names]
**Parallel**: [Yes / No / Partial]
**Goal**: [What's true when this phase completes]

---

## Task Execution Guides

### Task 1: [name]

**Prerequisites**: None / [Task N must be complete]

**Context to load**:

- Read: [specific files or docs]
- Prior output: [what dependency tasks produced — or "N/A" if first task]

**Key constraints**:

- [Constraint from analysis or user feedback]

**Start command**:

/workflow-analyze {task-name} [context summary for the session]


**This task produces**:

- [What downstream tasks will need from this task's output]

### Task 2: [name]

[Same structure]

### Task N: [name]

[Same structure]

---

## Milestones

### Milestone 1: [Name] (after Phase 1)

**Check condition**: [What to verify]
**If not met**: [What to do — re-plan, adjust scope, etc.]

### Milestone 2: [Name] (after Phase 2)

**Check condition**: [What to verify]
**If not met**: [What to do]

---

## Progress

| Task       | Status | Notes |
| ---------- | ------ | ----- |
| [Task 1]   | -      |       |
| [Task 2]   | -      |       |

Status: `-` pending / `>` in progress / `v` complete / `x` blocked

---

## Quick Reference: Execution Order

> This section is the at-a-glance summary. See details above.

1. [Task name] → `/workflow-analyze {name}`
2. [Task name] → `/workflow-analyze {name}`
3. [Task name] + [Task name] (parallel) → `/workflow-analyze {name}`
4. [Task name] → `/workflow-analyze {name}`

<!-- If all tasks are sequential, omit "(parallel)" markers -->
<!-- If some tasks are parallel, mark only those explicitly -->

Execution

Now start the task according to the guidelines above.

epic-plan

Epic Planning Command

User Input