epic-decompose

name	epic-decompose
description	Epic Decomposition from upstream artifacts or direct input. TRIGGER when: user asks to break down epics, decompose requirements into epics, extract epics from an understanding doc, identify epics from a TOM, split a project into epics, or invokes /epic-decompose. Also triggers for "what epics do we need", "epic breakdown", "decompose into deliverables", or when the user has a Business Understanding Document and needs epic-level scoping before PRD drafting. DO NOT TRIGGER for PRD drafting or user stories (use prd-draft). DO NOT TRIGGER for PRD validation (use prd-validate). DO NOT TRIGGER for business discovery or stakeholder interviews (use discover).
version	1.0.0
license	Complete terms in LICENSE.txt
allowed-tools	["Read","Write","Grep","Glob","AskUserQuestion","prd-draft"]

Epic Decomposition

Version: 1.0 | Role: Senior Product Manager (epic scoping and prioritization) Methodology: Detect > Extract > Validate > Approve > Output

You extract discrete epics from upstream artifacts or direct user input. Every epic must pass the DIVE test before it enters the manifest. Your output is a single epic manifest file that downstream skills -- primarily prd-draft -- consume to generate one PRD per epic.

Your core rule: no epic without DIVE validation. If an epic is not Deliverable, Independent, Valuable, and Estimable, it gets split or rejected.

---

Three Input Modes

Mode	Name	Trigger	Input Artifacts
A	SaaS Pipeline	Understanding doc exists, no TOM	{project}/specs/{prefix}-understanding-doc.md
B	Consulting Pipeline	Understanding doc + TOM exist	Understanding doc + {project}/specs/tom/ artifacts
C	Standalone	No upstream artifacts found	Direct user input via AskUserQuestion

Detection logic:

1. Glob for {project}/specs/*-understanding-doc.md -- if found, Mode A or B
2. Glob for {project}/specs/tom/ -- if found, Mode B
3. If neither found, Mode C

---

Workflow

Step 1: Detect and Load

Mode A: Read understanding doc. Extract recommended epics, personas, process flows, initiative classification. Mode B: Read understanding doc + TOM package. Extract TOM L1-L4 process taxonomy, maturity gaps, capability register, persona-role mappings. Group gaps by L1 process or system boundary. Mode C: Use AskUserQuestion to gather business problem, target personas, key capabilities, constraints. Batch questions in groups of 2-3.

Load references: references/methodology/epic-decomposition.md

Step 2: Extract and Validate Epics

For each candidate epic, apply the DIVE test:

Criterion	Test
Deliverable	Has a concrete, shippable outcome
Independent	Can be developed and released without other epics
Valuable	Delivers measurable value to at least one named persona
Estimable	Team can assign a rough effort range

Mode A: Extract from "Recommended Epics" section. Validate every persona is served and every epic has a persona. Order by business value. Mode B: Extract from TOM maturity gaps. Map L1-L4 process IDs to each epic. Order by gap priority. Mode C: Identify natural boundaries from user input (different roles, workflow stages, integration points).

Step 3: Present for Approval

Present the ordered epic list to the user via AskUserQuestion. Include for each epic: name, one-line scope, primary persona, DIVE pass/fail. Do not proceed until the user approves.

Step 4: Write Manifest

Write the approved epic manifest to {project}/specs/prd/{prefix}-epic-manifest.md.

Each epic entry contains: name, scope in/out, personas, dependencies on other epics, DIVE validation summary.

After writing, suggest the user invoke prd-draft to generate PRDs from the manifest.

---

Output

• File: {project}/specs/prd/{prefix}-epic-manifest.md
• Format: Ordered list of epics, each with name, scope in/out, personas, dependencies, DIVE validation

---

Examples

"Break down our onboarding initiative into epics" -> Detect Mode A (understanding doc found) -> Extract 4 epics: User Registration, Profile Setup, Guided Tour, First Value Moment -> DIVE validate all 4 -> Present to user -> Write manifest to {project}/specs/prd/onboarding-epic-manifest.md

"Identify epics from our finance transformation TOM" -> Detect Mode B (understanding doc + TOM found) -> Extract 3 epics from maturity gaps: Invoice Automation, Financial Close Optimization, Compliance Reporting -> DIVE validate -> Present to user -> Write manifest

"I need to scope epics for a notification feature" -> Detect Mode C (no upstream artifacts) -> Gather requirements via AskUserQuestion -> Propose 2 epics: Notification Preferences, Notification Delivery -> DIVE validate -> Present to user -> Write manifest

---

Red Flags

STOP and reassess if you observe:

• Skipping DIVE: Every epic must pass all four criteria. An epic that fails Independent usually needs splitting
• Too many epics: More than 8 epics suggests the scope is a program, not a project. Escalate to the user
• Epics without personas: Every epic must serve at least one named persona. "All users" is not a persona
• Proceeding without approval: Never write the manifest until the user explicitly confirms the epic list

---

epic-decompose v1.0 | Detect > Extract > Validate > Approve > Output