Skip to content

Dimon94/brainstorming-only

Repository files navigation

Brainstorming-Only

中文说明

npm version License: MIT Node.js >=16 GitHub stars

Two bounded AI agent skills: focused brainstorming and office-hours-style product/business-model diagnosis, both stopping before specs or implementation.

Supports both Codex and Claude skill directories from one shared skill bundle.

This project is a two-skill package. $brainstorming-only 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. $office-hours-only distills the product-diagnostic core of office-hours from gstack: demand reality, status quo, specific users, narrow wedges, observation, future-fit, premise challenge, and alternatives. It adds an explicit business-model chain while staying independent from the gstack runtime and downstream workflow machinery.

When brainstorming is tied to a local project, durable terms and hard-to-reverse decisions are captured through project context docs: glossary-only CONTEXT.md files and sparse ADRs under docs/adr/. It does not keep a separate session journal, hidden local cache, or recovery directory.

Quick Install

Install both Codex and Claude skill copies with one command:

npx --yes brainstorming-only --both

Codex-only:

npx --yes brainstorming-only --codex

Claude-only:

npx --yes brainstorming-only --claude

The installer writes to:

${CODEX_HOME:-~/.codex}/skills/brainstorming-only
${CODEX_HOME:-~/.codex}/skills/office-hours-only
${CLAUDE_HOME:-~/.claude}/skills/brainstorming-only
${CLAUDE_HOME:-~/.claude}/skills/office-hours-only

Global install also works:

npm install -g brainstorming-only
brainstorming-only --codex

Requirements:

  • Node.js 16 or newer.
  • A host that supports local skill directories.
  • Write access to ~/.codex/skills or ~/.claude/skills.

Use

In an AI agent that supports skills:

Use $brainstorming-only to compare these product directions and help me choose one.
Use $office-hours-only to pressure-test this product and business model.

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.

When the discussion resolves durable project language, the skill updates the applicable CONTEXT.md inline. ADRs are offered only when a decision is hard to reverse, surprising without context, and the result of a real trade-off.

What It Adds

  • Discussion-only brainstorming with a hard boundary before specs, plans, scaffolds, commits, or PRs.
  • Two postures: general brainstorming and builder mode.
  • A clean handoff to $office-hours-only for product viability, demand, narrowest wedge, observation, pricing, distribution, economics, and commercial durability.
  • 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.
  • Unified A/B/C option output for both blocking pauses and terminal convergence, always with one recommended option and the reason it wins.
  • Project context docs as the only persistence surface: glossary-only CONTEXT.md updates for resolved language and sparse ADRs for durable trade-off decisions.

Skill Boundary

  • General brainstorming - clarify a fuzzy idea, compare directions, and converge on a decision.
  • Builder mode - explore the most delightful version of a side project, open source idea, learning project, demo, or hackathon concept.
  • Office hours - use $office-hours-only to trace a specific user's pain through the product wedge, observed demand, buyer, value capture, distribution, economics, and future-fit.

Structured Choices

The skill uses one visible option format for blocking choice pauses and terminal convergence. 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 response still uses A/B/C options, marks one option as recommended, and explains why that recommendation wins.

The choice protocol is only a host-rendering adapter. Upstream brainstorming flow decides the recommendation, reliability note, and whether the choice is a blocking pause or terminal convergence before the protocol renders it.

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 = true

Without 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 choice rendering contract and host mapping live in brainstorming-only/references/user-choice-output-protocol.md.

Privacy And Local Data

The package no longer creates local recovery journals or hidden cache directories. For local project discussions, durable knowledge is written only to project-owned context docs when it has been resolved: CONTEXT.md, pre-existing CONTEXT-MAP.md, or sparse ADRs under docs/adr/.

No gstack telemetry, analytics, or runtime commands are required for this package.

Prior Art And Attribution

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 $brainstorming skill, 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 diagnostic lens now isolated in $office-hours-only. This package borrows those ideas respectfully while staying independent from the gstack runtime, telemetry, and downstream workflow machinery.

Recommended Skill Suites

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/.

Project Signals

Current public repository signals as of 2026-05-22:

Star History Chart

Verify A Local Checkout

npm test
npm run pack:check
node scripts/install.js --codex

If you later want implementation or planning, treat that as a separate request after the brainstorming session ends.

What's Included

  • brainstorming-only/SKILL.md - the skill instructions.
  • brainstorming-only/references/brainstorming-method.md - detailed brainstorming workflow, postures, pressure tests, and output shape.
  • brainstorming-only/references/project-context-docs.md - CONTEXT.md and ADR persistence rules.
  • brainstorming-only/references/recommendation-reliability.md - second-sample and roundtable recommendation checks.
  • brainstorming-only/references/user-choice-output-protocol.md - host-specific structured choice protocol.
  • brainstorming-only/agents/openai.yaml - optional UI metadata for compatible skill lists.
  • office-hours-only/SKILL.md - the standalone product and business-model diagnostic skill.
  • office-hours-only/references/office-hours-method.md - forcing questions, evidence ladder, product-to-business chain, premise challenge, and assignment.
  • office-hours-only/agents/openai.yaml - optional UI metadata for the office-hours skill.
  • scripts/install.js - npm installer that copies both skills into CODEX_HOME and CLAUDE_HOME.

Documentation

License

MIT

About

Standalone AI agent skill for brainstorming without specs, plans, commits, scaffolding, or implementation.

Resources

License

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors