Standalone AI agent skill for focused brainstorming without drifting into specs, plans, commits, scaffolding, or implementation.
Supports both Codex and Claude skill directories from one shared SKILL.md.
This project is a single-skill package. It is a focused, standalone homage to
the original $brainstorming skill
from Superpowers: it keeps the
collaborative clarification, option generation, and trade-off analysis, while
deliberately stopping before any downstream workflow begins. It also adds an
office-hours-inspired
product diagnostic posture from gstack
for sharper "is this worth building?" conversations, while staying independent
from gstack telemetry, design docs, commits, and downstream workflow machinery.
For long sessions, it keeps a small local recovery journal under
~/.brainstorming/. The journal uses the same project slug shape as
office-hours
without depending on the gstack runtime, so compressed context can recover the
latest first-hand decisions without writing into the project workspace.
Install both Codex and Claude skill copies with one command:
npx --yes brainstorming-only --bothCodex-only:
npx --yes brainstorming-only --codexClaude-only:
npx --yes brainstorming-only --claudeThe installer writes to:
${CODEX_HOME:-~/.codex}/skills/brainstorming-only
${CLAUDE_HOME:-~/.claude}/skills/brainstorming-onlyGlobal install also works:
npm install -g brainstorming-only
brainstorming-only --codexRequirements:
- Node.js 16 or newer.
- A host that supports local skill directories.
- Write access to
~/.codex/skillsor~/.claude/skills.
In an AI agent that supports skills:
Use $brainstorming-only to compare these product directions and help me choose one.
The skill is intentionally discussion-only. It may help frame the topic, choose a brainstorming posture, ask targeted questions, generate options, stress-test trade-offs, and summarize the decision. It must not create specs, write implementation plans, scaffold files, commit changes, open PRs, or transition into another workflow.
During longer conversations, the skill checkpoints the session after every 10 effective question-answer pairs and immediately after key decisions. Each checkpoint is written to:
~/.brainstorming/projects/<project-slug>/sessions/<timestamp>-<topic-slug>/brainstorming.md
The companion meta.json tracks checkpoint_count, last_checkpoint_at,
qa_since_checkpoint, status, and the active / latest recovery pointers.
- Discussion-only brainstorming with a hard boundary before specs, plans, scaffolds, commits, or PRs.
- Three postures: general brainstorming, product diagnostic, and builder mode.
- Adversarial clarity: the skill names weak assumptions, missing evidence, and failure modes instead of cheaply agreeing.
- Structured decision pauses that use the host's native choice UI when available.
- Outcome open options that carry non-blocking A/B/C alternatives inside the final brainstorming summary.
- Context recovery journal under
~/.brainstorming/, with checkpoints every 10 effective question-answer pairs and immediately after key decisions. - Standalone project slug logic inspired by
office-hours, without depending on gstack runtime files, commands, telemetry, or caches.
- General brainstorming - clarify a fuzzy idea, compare directions, and converge on a decision.
- Product diagnostic - pressure-test demand, current workarounds, concrete users, narrowest wedge, observation, and future-fit.
- Builder mode - explore the most delightful version of a side project, open source idea, learning project, demo, or hackathon concept.
The skill separates blocking choice pauses from terminal outcomes. If the next
reasoning step requires the user to choose, the skill uses the host's native
choice UI when available. If the recommendation is already clear, the final
Brainstorming outcome can carry non-blocking A/B/C Open options instead of
stopping for a separate choice.
For Codex blocking choices, that means request_user_input when it is listed in
the current turn's tools. Codex can expose that tool in Default mode when this is
configured in ~/.codex/config.toml:
[features]
default_mode_request_user_input = trueWithout that Codex config, the skill falls back to a fixed Markdown A/B/C choice
block and waits for the user to reply with A, B, or C. The Markdown block is
only a fallback; it is not native Codex option UI. For Claude Code, native choice
means MCP elicitation unless the official host protocol changes. For gstack-style
hosts, that means a real AskUserQuestion tool when available.
To enable the Codex setting directly, send Codex this prompt:
Please enable Codex Default-mode choice popups by adding [features] default_mode_request_user_input = true to ~/.codex/config.toml, preserving all existing settings.
The detailed host mapping lives in
brainstorming-only/references/user-choice-output-protocol.md.
The session journal is local-only. It writes under ~/.brainstorming/, not the
current project directory, and it is meant for recovery after context
compression. The skill instructs agents not to record credentials, tokens,
private keys, or sensitive personal data. If sensitive material appears during a
conversation, the journal should capture only the decision-relevant constraint
and redact the secret itself.
No gstack telemetry, analytics, or runtime commands are required for this package.
brainstorming-only exists because other open projects made the underlying
workflow patterns concrete. This package is independently maintained and is not
affiliated with, sponsored by, or endorsed by those projects or their
maintainers. The projects below remain under their own maintainers and licenses.
We cite them directly here with respect:
- Superpowers, especially its
$brainstormingskill, shaped the collaborative clarification, alternative generation, trade-off framing, and explicit pre-implementation boundary that this package honors in a narrower discussion-only form. - gstack, especially its
office-hours, inspired the sharper product diagnostic posture and the project-slug-shaped recovery journal. This package borrows those ideas respectfully while staying independent from the gstack runtime, telemetry, and downstream workflow machinery.
If you are choosing an agent-skill workflow, this package is the intentionally small option: use it when you want to think clearly without starting a plan or implementation loop. For broader workflows, we recommend these adjacent skill suites:
| Skill suite | Best fit | Work situations |
|---|---|---|
| Superpowers | Developers and teams who want a broad, composable agentic software-development methodology across coding agents. | Use it when you want the agent to follow disciplined workflows for idea shaping, planning, test-driven implementation, debugging, review, and shipping rather than relying on ad hoc prompts. |
| gstack | Founders, product builders, and high-agency teams who want stronger product judgment, design review, execution review, QA, ship, and retro loops. | Use it when the hard part is deciding what is worth building, improving taste and ambition, turning product intent into a stronger plan, and carrying that plan through a full delivery lifecycle. |
| cc-devflow (same maintainer) | Maintainers and agent-assisted engineering teams who want a small, explicit repo workflow with one roadmap entry point, feature and bug-investigation loops, verification gates, PR review, and main-branch parity. | Use it when you need repeatable project execution: roadmap to next work item, plan or investigation, implementation, fresh verification, PR creation, review-first landing, and durable project artifacts under devflow/. |
Current public repository signals as of 2026-05-22:
- GitHub repository: Dimon94/brainstorming-only
- Stars: 0
- Watchers: 0
- npm package: brainstorming-only
npm test
npm run pack:check
node scripts/install.js --codexIf you later want implementation or planning, treat that as a separate request after the brainstorming session ends.
brainstorming-only/SKILL.md- the skill instructions.brainstorming-only/references/user-choice-output-protocol.md- host-specific structured choice protocol.brainstorming-only/scripts/journal.js- local session journal helper.brainstorming-only/agents/openai.yaml- optional UI metadata for compatible skill lists.scripts/install.js- npm installer that copies the skill intoCODEX_HOMEandCLAUDE_HOME.
MIT