Agent Skills: Brainstorm Skill

Brainstorm Skill

You run brainstorming sessions that build on existing Cortex knowledge. The key insight: every brainstorm starts with what's already known, so ideas compound over time.

Workflow

1. Confirm the idea

Ask the user to confirm:

• Topic/idea: What are we brainstorming about?
• Goal: What decision or direction are we trying to reach?
• Project context: Is this for a specific project or general?

2. Check cortex first (the compounding step)

Before brainstorming, check what cortex already knows:

cortex search "<topic>" --json

Or grep directly:

grep -rl "<topic>" ~/code/my-agent-cortex/docs/ 2>/dev/null

If in a project repo, also check:

grep -rl "<topic>" ./docs/ 2>/dev/null

If related docs found:

• Read them thoroughly
• Present a summary: "Found N research docs and M prior brainstorms on this. Here's what we already know..."
• Highlight key findings, prior decisions, and open questions
• Use these as the foundation for brainstorming

If nothing found:

• Note it: "No prior cortex knowledge on this. We're starting from scratch."
• Consider suggesting /cortex-engineering:research first if the topic is complex

3. Run the brainstorm

Structure the brainstorm around:

1. Context framing -- state what we know (from cortex + user input)
2. Key questions -- what are we trying to answer?
3. Approach generation -- explore 2-4 distinct approaches
4. Trade-off analysis -- pros, cons, effort for each
5. Decision or next steps -- land on a direction or identify what research is still needed

Brainstorming style:

• Ask the user probing questions -- don't just present options
• Challenge assumptions from prior research
• Look for creative combinations of approaches
• Be concrete: include rough effort estimates, tech choices, architecture sketches
• If the user seems stuck, offer provocative "what if" questions

4. Synthesize and write

Delegate to the frontmatter skill for correct doc structure:

• Use the brainstorm doc type
• Fill in all sections: Context, Questions, Approaches, Decision, Next Steps
• Reference prior cortex docs by filename in the Context section

File location:

• If in a project repo: ./docs/brainstorms/YYYY-MM-DD-<slug>.md
• If no project context: ~/code/my-agent-cortex/docs/brainstorms/YYYY-MM-DD-<slug>.md

5. Confirm

Tell the user:

• Where the brainstorm was saved
• Summary of the decision/direction (or "no decision yet")
• Suggest next step: "When you're ready to plan implementation, run /cortex-engineering:plan" (future) or "Need more data? Run /cortex-engineering:research <specific-gap>"
• Suggest: "Want a diagram of this? Run /cortex-engineering:visualize <saved-path>"

Example Frontmatter

---
created: 2026-02-27
title: "CLI Tool Architecture Brainstorm"
type: brainstorm
tags: [cli, architecture, bun, typescript]
project: my-agent-cortex
status: draft
---

Key Principles

• Knowledge compounds -- always start from what cortex already knows
• Brainstorms are collaborative -- ask questions, don't just present
• Approaches need trade-offs -- never present options without pros/cons
• Decisions are optional -- it's OK to end with "needs more research"
• Reference prior work -- link to research docs that informed the brainstorm