Brainstorming
Use this skill to decide WHAT to build before deciding HOW.
Default Interaction Contract
This skill defaults to collaborative back-and-forth.
The user should never need to say special phrases like "work back and forth with me" to get interactive brainstorming behavior.
Do not skip directly to a brainstorm doc or a recommended approach unless the user explicitly asks for a one-shot draft.
One-Shot Exception
If the user explicitly asks for a fast draft:
- say you are proceeding in one-shot mode
- make assumptions explicit
- include an
Assumptions to Validatesection in the brainstorm doc - still pause if there is a critical blocker or major unresolved tradeoff
Note: current year 2026. Use when dating brainstorm docs.
When to Use
Use when:
- requirements are vague/ambiguous
- multiple valid approaches exist
- tradeoffs need user alignment
- success criteria are unclear
Skip when:
- requirements are explicit and detailed
- straightforward bug fix with known path
If skip applies, suggest moving to loops-plan.
Workflow
Phase 1: Understand the Idea
Run a light repo scan for related patterns.
- look for similar features
- capture conventions from
AGENTS.md
Run collaborative dialogue.
- ask one question at a time
- prefer multiple-choice when natural
- start broad then narrow
- validate assumptions explicitly
- ask success criteria early
Key topics:
- purpose/problem
- users/context
- constraints/dependencies
- success criteria
- edge/error cases
- existing patterns to copy
Before moving to approaches, present a short understanding summary in the chat:
- problem to solve
- target users/context
- constraints
- success criteria
- open questions
- explicit assumptions
Ask the user to confirm or adjust this summary before moving to approaches, unless one-shot mode was explicitly requested.
Exit when idea is clear or user says proceed.
Phase 2: Explore Approaches
Propose 2-3 concrete approaches.
For each approach include:
- short description (2-3 sentences)
- pros
- cons
- best-fit context
Lead with recommendation. Explain tradeoff. Apply YAGNI.
Phase 3: Capture Design
Write file:
docs/brainstorms/YYYY-MM-DD-<topic>-brainstorm.md
Ensure docs/brainstorms/ exists.
Template:
---
date: YYYY-MM-DD
topic: <kebab-case-topic>
---
# <Topic Title>
## What We're Building
[1-2 short paragraphs]
## Why This Approach
[brief tradeoff summary]
## Key Decisions
- [Decision 1]: [Rationale]
- [Decision 2]: [Rationale]
## Open Questions
- [Unresolved items for planning]
## Assumptions to Validate
- [Explicit assumptions that still need confirmation]
## Next Steps
- Move to `loops-plan` for implementation planning
Phase 4: Optional Inline Refinement (Self-Contained)
If user wants refinement, run an inline review pass in this same skill:
- Assess clarity, completeness, specificity, YAGNI
- Auto-fix minor wording/formatting issues
- Ask before substantive scope or meaning changes
- Update the same brainstorm file directly
After up to 2 passes, recommend moving on.
Phase 5: Handoff
Ask user what next:
- refine brainstorm now (inline)
- proceed to
loops-plan - done for now
Output Summary
Return concise closeout:
Brainstorm complete.
Document: docs/brainstorms/YYYY-MM-DD-<topic>-brainstorm.md
Key decisions:
- ...
- ...
Next: run loops-plan when ready.
Rules
- stay on WHAT, not HOW
- one question at a time
- default to simple path (YAGNI)
- keep sections concise (target 200-300 words max)
- pause for alignment after major section
- do not implement code in this skill
Anti-Patterns
Avoid:
- many questions at once
- implementation deep dives
- unvalidated assumptions
- over-designed speculative scope
- long docs with low signal
Planning Integration
When brainstorm doc exists, loops-plan should detect and use it.
If relevant brainstorm found, skip repeated ideation and start from decisions already captured.