Brainstorming
Skill fusengine/agents/plugins/ai-pilot/skills/brainstorming
Use when user requests creative work - creating features, building components, adding functionality, or modifying behavior. Activates BEFORE APEX Analyze phase to refine requirements through structured questioning.From its SKILL.md
npx -y skills add fusengine/agents --skill brainstormingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 22 stars22 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 1 command, including `git log --oneline -20`.
SKILL.md
4.1 KB, 896 tokens by cl100k_base, as published. Nobody here has run it
Brainstorming Skill
Design-first approach: no code before design approval.
Agent Workflow (MANDATORY)
Before ANY brainstorming session, use TeamCreate to spawn agents:
- fuse-ai-pilot:explore-codebase - Understand project context, patterns, constraints
- fuse-ai-pilot:research-expert - Fetch best practices and documentation
After design approval, transition to APEX Analyze phase, passing along the explore-codebase and research-expert findings in the handoff (not just the design doc). APEX Analyze re-runs those same agents by default — carrying forward what was already gathered here avoids re-doing the same research twice.
Overview
| Scenario | Action |
|---|---|
| New feature | Full brainstorming (6 steps) |
| Major change | Full brainstorming (6 steps) |
| Component creation | Full brainstorming (6 steps) |
| Trivial fix (1-3 lines) | Skip brainstorming, go to APEX |
| Simple rename/typo | Skip brainstorming, go to APEX |
6-Step Process
Step 1: Explore Project Context
Gather context before asking questions:
git log --oneline -20- Recent changes and direction- Existing code patterns and conventions
- Related documentation and prior decisions
- Tech stack and dependency constraints
Step 2: Ask Clarifying Questions (ONE AT A TIME)
Ask focused questions sequentially. Wait for each answer before the next.
Categories: purpose, constraints, success criteria, users, integrations.
See workflow.md for question categories
Step 3: Diverge, Then Converge to 2-3 Approaches
Generate ≥6-8 distinct approaches via a named technique (SCAMPER / reverse-brainstorming / analogies), judgment suspended — not 3 sizes of the same idea. Only then converge to 2-3 with trade-offs in table format:
| Approach | Pros | Cons | Recommendation |
|---|---|---|---|
| Option A | ... | ... | ... |
| Option B | ... | ... | ... |
| Option C | ... | ... | ... |
Always include a clear recommendation with rationale.
Step 4: Present Design for Approval
Break design into digestible sections:
- Architecture overview
- Key components and responsibilities
- Data flow and state management
- Edge cases and error handling
Wait for explicit user approval before proceeding.
Step 5: Save Design Document
Save approved design to: docs/plans/YYYY-MM-DD-<topic>-design.md
Step 6: Transition to APEX
Hand off to APEX Analyze phase with the approved design as input.
Reference Guide
| Topic | Reference | When to Consult |
|---|---|---|
| Workflow | workflow.md | Question categories, proposal format, design template |
| Anti-Patterns | anti-patterns.md | Catching rationalizations to skip brainstorming |
Quick Reference
1. Explore → git log, codebase, docs (agents in parallel)
2. Question → ONE AT A TIME, wait for answers
3. Diverge → 6-8 options (named technique), converge to 2-3 with trade-offs table
4. Design → Present sections, get approval
5. Save → docs/plans/YYYY-MM-DD-<topic>-design.md
6. Handoff → APEX Analyze with approved design + prior research (no re-research)
Critical Rules
- NEVER write code before design approval - Design first, always
- Ask questions ONE AT A TIME - Never dump a list of 10 questions
- Always propose alternatives - Minimum 2 approaches with trade-offs
- Save the design doc - Creates audit trail and shared reference
- Get explicit approval - "Looks good" or similar before proceeding
What ships with it: 2 files
10.7 KB alongside SKILL.md
references/
- anti-patterns.md4.4 KB
- workflow.md6.3 KB