Design

/design

Create PRDs through structured decision-tree interviews and collaborative dialogue.

Guiding Principles

• No implementation until PRD is approved. No Plan Mode — this skill operates in normal mode. EnterPlanMode/ExitPlanMode restrict Write/Edit tools and break the workflow.
• The session-start hook injects a metadata block into the session context with epic-id, epic-slug, and output-target. Use these values verbatim — do NOT re-derive, re-extract, or generate alternatives.
• Do NOT invoke any implementation skill, write any code, scaffold any project, or take any implementation action until you have presented a PRD and the user has approved it. This applies to EVERY project regardless of perceived simplicity.
• Every feature gets a PRD. A config change, a single-function utility, a rename — all of them. Short is fine; skipping is not.
• The user frames the problem. Their words drive the design. Ask before exploring; listen before loading context.
• Decision tree before document. Walk every branch, sweep for gray areas, then write. The PRD is the output, not the process.
• Scope is protected by default. New capabilities get deferred unless the user explicitly pulls them in.

Phase 0: Pre-Execute

1. Problem-First Question

Before exploring the codebase, ask the user what they are trying to solve.

Do NOT:

• Explore the codebase yet
• Load project context yet

Wait for the user's response. Their framing drives the entire design — do not proceed until they answer.

2. Express Path Check

If the user's response points to an existing PRD, spec, or requirements document (not a .beastmode/artifacts/design/ file):

1. Read the document
2. Skip decision tree walk in execute
3. Jump directly to "Gray Areas" (Execute step 2) with the doc as input

3. Existing Design Check

If a prior PRD exists for the same topic (matching feature name):

• Ask: "Found existing PRD for this topic. What do you want to do?"
• Options: Update existing / View first / Start fresh

Phase 1: Execute

1. Walk the Decision Tree

Interview the user about every aspect of this feature. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one.

Rules:

1. Ask questions one at a time
2. For each question, provide your recommended answer
3. If a question can be answered by exploring the codebase, explore the codebase instead of asking
4. If a question requires research (unfamiliar technology, external APIs, best practices), spawn an Explore agent as the researcher. It receives the research topic and returns findings with sources. Save findings to .beastmode/artifacts/research/YYYY-MM-DD-.md
5. Honor prior decisions from pre-execute checks — don't re-ask settled questions
6. Scope guardrail: new capabilities get deferred

"That sounds like its own feature — I'll note it as a deferred idea. Back to the current branch."

1. Track deferred ideas internally
2. Continue until all branches of the decision tree are resolved

2. Gray Area Sweep

Second pass to catch big-picture blind spots the decision tree may have missed.

Maintain a session-scoped list of resolved gray areas. Never re-present a resolved area.

Loop:

1. Analyze remaining ambiguity (excluding resolved areas), rank by ambiguity level
2. If 0 gray areas remain → sweep complete, proceed to step 3
3. Present the single most ambiguous gray area via AskUserQuestion with multiSelect: false

• Options: relevant resolution choices for that gray area (recommendation first)
• Annotate with codebase context when relevant

1. If the user responds via Other with a bail-out intent (skip, done, move on, etc.) → exit loop immediately
2. Resolve the gray area based on the selected option or freeform input

• Scope guardrail: defer new capabilities

1. Add to session-scoped resolved list
2. Loop back to step 1

3. Iterate Until Ready for Validation

• Go back and clarify as needed
• Keep YAGNI in mind — remove unnecessary features
• PRD is ready when decision tree + gray areas are all resolved

Phase 2: Validate

1. Completeness Check

Verify PRD covers all required sections:

• [ ] Problem Statement
• [ ] Solution
• [ ] User Stories (3+ items)
• [ ] Implementation Decisions
• [ ] Testing Decisions
• [ ] Out of Scope
• [ ] Further Notes (or "none")
• [ ] Deferred ideas captured (or "none")

If missing sections, go back to Execute phase.

2. Anti-Pattern Check

If the PRD produced fewer than 3 user stories, print:

"This is a lightweight PRD — that's fine. Even simple features benefit from explicit approval to prevent wasted implementation."

Do NOT skip approval. Short PRDs still need approval.

3. Executive Summary

Before asking for approval, present a consolidated executive summary so the user can review the full picture in one place.

Print:

### Executive Summary

**Problem**: [one-sentence problem statement]

**Solution**: [one-sentence solution summary]

**Key Decisions:**

| Decision | Choice |
|----------|--------|
| [decision 1] | [choice] |
| [decision 2] | [choice] |
| ... | ... |

**User Stories:** [count] stories covering [summary of scope]

Render this from the decisions and stories gathered during the execute phase. Do NOT ask new questions — this is a read-only summary of what was already discussed.

4. PRD Approval

Ask: "Does this PRD look complete? Ready to document?"

Options:

• Yes, document it
• No, let's revise [specify what]

Wait for user response before continuing.

Phase 3: Checkpoint

0. Derive Epic Name

Derive a NEW human-readable name from the problem statement and solution discussed during the interview:

• epic-name: A short, human-readable title (e.g. "Dashboard Redesign for Log Events"). Not slugified — natural language, title case. The CLI slugifies this automatically for the epic slug.

• Log: "Derived name: ``"

1. Write PRD

Save to the path specified by output-target in the session metadata block.

In the PRD frontmatter:

• epic-id: from session metadata (verbatim)
• epic-slug: from session metadata (verbatim)
• epic-name: the human-readable name from step 0

Use the PRD template from the Reference section below.

2. Commit and Handoff

Commit all work to the feature branch:

git add -A
git commit -m "design(): checkpoint"

Print:

Next: beastmode plan 

STOP. No additional output.

Reference

PRD Template

---
phase: design
epic-id: 
epic-slug: 
epic-name: 
---

## Problem Statement

[The problem from the user's perspective]

## Solution

[The solution from the user's perspective]

## User Stories

[Numbered list of user stories in format: As an , I want a , so that ]

## Implementation Decisions

[Flat list of implementation decisions made during the interview. Include:
- Interfaces that will be modified
- Technical clarifications
- Architectural decisions
- Schema changes, API contracts, specific interactions

Do NOT include specific file paths or code snippets — they may become outdated.]

## Testing Decisions

[Include:
- What makes a good test for this feature
- Prior art for tests (similar test patterns in the codebase)]

## Out of Scope

[Things explicitly excluded from this PRD]

## Further Notes

[Additional context, or "None"]

## Deferred Ideas

[Ideas that came up during the interview but were deferred as separate features, or "None"]

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: BugRoger
• Source: BugRoger/beastmode
• License: MIT

Install and usage instructions live in the source repository linked above.