Take any idea from rough concept to dev-ready deliverables — PRD, phased plan, and dependency-ordered GitHub issues — using structured AI-assisted skills.
Scope creep, vague requirements, and dev misalignment kill projects before they start. Brainstormer is a consulting toolkit that forces clarity before a single line of code is written. The output is a structured handoff package that any dev team can pick up and run with.
Consultants, product owners, founders, and technical leads — working with any tech stack. Use this per engagement to guide a client from "I have an idea" to "here are the issues, start building."
- PRD — Comprehensive requirements document, submitted as a GitHub issue
- Phased plan — Vertical-slice implementation plan saved to
./plans/ - GitHub issues — Dependency-ordered, independently-grabbable work items with HITL/AFK classification
claude plugin install <repo-url>All 10 skills become available via / commands in any project.
Copy skills/ into your project's .claude/skills/ or your global ~/.claude/skills/:
cp -R skills/* ~/.claude/skills/Clone and work directly in the repo:
git clone <repo-url>
cd brainstormer
# Run /brainstorm to start the full workflow- An AI code assistant (Claude Code, Cursor, GitHub Copilot, etc.)
- GitHub CLI (
gh) — authenticated (required for/blueprintand/dispatch) - A GitHub repository for the engagement
| Skill | Command | Purpose |
|---|---|---|
| Brainstorm | /brainstorm |
Full orchestrated workflow — discovery through issue creation |
| Ask | /ask |
Discovery interview — pressure-test the idea, surface constraints |
| Blueprint | /blueprint |
Structured interview to produce a PRD (GitHub issue) |
| Carve | /carve |
Break PRD into phased vertical slices (./plans/) |
| Dispatch | /dispatch |
Create dependency-ordered GitHub issues from PRD |
| TDD | /tdd |
Red-green-refactor workflow with GitHub issue + plan integration |
| Lean | /lean |
Check any decision for over-engineering |
| Ship | /ship |
Lint, type-check, commit, and push in one flow |
| LLM Council | /llm-council |
Multi-perspective advisory council for strategic decisions |
| Improve CLAUDE.md | /improve-claude-md |
Audit and optimize CLAUDE.md files with conditional importance tags |
Start here with a rough idea. The assistant runs a structured interview covering the problem, users, business model, constraints, and domain-specific concerns. Walk away with a clear summary of decisions and open questions.
Formalize the discovery output into a Product Requirements Document. The assistant interviews you, sketches system components, and submits the PRD as a GitHub issue with user stories, implementation decisions, and validation strategy.
Break the PRD into tracer-bullet phases. Each phase is a thin vertical slice cutting through the full system end-to-end. Output is a markdown plan in ./plans/.
Convert the plan into GitHub issues. Each issue is a vertical slice classified as AFK (autonomous) or HITL (needs human input), created in dependency order with blocker links.
Pick up an issue and implement it using red-green-refactor. Pass a GitHub issue number (/tdd #123) to pull in scope and acceptance criteria automatically, or start from scratch. The skill reads plans from ./plans/ for architectural context.
Use
/leanat any point to gut-check a decision against MVP principles.
Or run /brainstorm to be guided through all phases in sequence.
Brainstormer works with any AI code assistant that reads instruction files:
- Claude Code reads
CLAUDE.md(symlinked toAGENTS.md) - GitHub Copilot, Cursor, and others read
AGENTS.md - Any agent can fetch
llms.txtfor a machine-readable index of all skills and references
Point your agent at llms.txt in the repo root for programmatic discovery of all skills, templates, and references. Each entry links to the relevant file with a one-line description.
# Got a rough idea for a client portal? Start with discovery.
/ask
> "We need a portal where clients can upload documents and track project status"
> Assistant probes: who are the clients, how many, what document types,
> sensitivity level, existing systems, timeline, budget...
# Formalize into requirements
/blueprint
> Assistant interviews, creates GitHub issue #1 with full PRD
# Break into phases
/carve
> Creates ./plans/client-portal.md with vertical slices
# Create trackable issues
/dispatch
> "The PRD is issue #1"
> Creates issues #2-#8 in dependency order, ready for dev handoff
# Implement with TDD
/tdd #2
> Fetches issue #2, reads ./plans/client-portal.md for context,
> walks through red-green-refactor cycles with you
These skills power Brainstormer Consulting — a 5-agent AI company built on the Agent Companies protocol. Deploy on Paperclip for multi-agent orchestration with org charts, budgets, and approval gates.
Brainstormer is technology-agnostic at the planning layer. After the tech stack is decided, add project-specific skills per engagement (e.g., deployment procedures, coding standards).
Each skill is conversational — the assistant will ask questions and iterate with you before taking action.