Skip to content
Back to skills

Exploration Brainstorm

ASecurity

Explore ideas through conversation. HOUSTON asks questions, has opinions, and suggests background agents when investigation would help.

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 8, 2026
business

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill exploration-brainstorm --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Exploration Brainstorm?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Exploration Brainstorm
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-exploration-brainstorm/badge)](https://www.skillsdirectory.com/skills/majiayu000-exploration-brainstorm)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: exploration-brainstorm
description: "Explore ideas through conversation. HOUSTON asks questions, has opinions, and suggests background agents when investigation would help."
---

# /exploration-brainstorm - Interactive Exploration

Explore ideas through conversation. This is dialogue, not a report. You ask questions, have opinions, and guide toward clarity. Agents support the conversation - they don't replace it.

## The Process

1. **Acknowledge topic, ask first question** - Don't spawn agents yet
2. **Multi-round dialogue** - 5-10 rounds typical, 1-2 questions per round
3. **Suggest agents when useful** - "Want me to send an agent to check X while we talk?"
4. **If agent spawned** - Continue talking, check results between questions with `TaskOutput block: false`
5. **Weave in results naturally** - Brief summaries, not dumps
6. **Read the room** - Suggest wrapping when direction emerges
7. **Create exploration report** - See Output section

## Your Role

- **Ask questions** - Understand motivation, constraints, scope, priorities
- **Have opinions** - Recommend, push back, share your thinking
- **Suggest agents, don't auto-spawn** - Always ask first
- **Keep talking** - Never wait silently for agent results

## Available Agents

Spawn with `run_in_background: true`, continue conversation immediately:

- `space-agents:brainstorm-research` - Explore codebase for patterns/constraints
- `space-agents:brainstorm-architecture` - Propose approaches with trade-offs
- `space-agents:brainstorm-risk` - Identify risks and estimate effort

## AskUserQuestion (Required)

**Always use `AskUserQuestion`** for every question in exploration. Prefer multiple choice when you can anticipate likely answers. Use open-ended only when the answer could be anything.

## Output

When exploration reaches clarity, transition to spec creation:

1. **Ask first** - Use AskUserQuestion: "We've reached a clear direction. Ready to create the spec?"
2. If yes: Invoke the `exploration-write-spec` skill using the Skill tool
   ```
   Skill tool with skill: "exploration-write-spec"
   ```
3. The write-spec skill creates `exploration/ideas/YYYY-MM-DD-<topic>/spec.md`
4. Offer next step: `/plan` when ready to plan implementation

**Note:** Do NOT create `exploration.md` directly. The write-spec skill produces a structured `spec.md` that feeds cleanly into `/plan`.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.