{"id":50402059,"url":"https://github.com/winchxyz/idea-to-build","last_synced_at":"2026-06-05T05:00:49.348Z","repository":{"id":361241727,"uuid":"1253624941","full_name":"winchxyz/idea-to-build","owner":"winchxyz","description":"Turn a raw idea into a plan you can build. A multi-agent brainstorming methodology — isolated critique, fact-checking, and a build handoff to Claude Code.","archived":false,"fork":false,"pushed_at":"2026-06-03T20:49:21.000Z","size":5132,"stargazers_count":3,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-06-04T04:11:43.360Z","etag":null,"topics":["ai-agents","brainstorming","claude","claude-code","cowork-plugin","mcp","methodology","multi-agent","open-source","prompt-engineering"],"latest_commit_sha":null,"homepage":null,"language":"Shell","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/winchxyz.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-29T16:48:20.000Z","updated_at":"2026-06-03T20:49:25.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/winchxyz/idea-to-build","commit_stats":null,"previous_names":["winchxyz/idea-to-brainstorm","winchxyz/idea-to-build"],"tags_count":11,"template":false,"template_full_name":null,"purl":"pkg:github/winchxyz/idea-to-build","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/winchxyz%2Fidea-to-build","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/winchxyz%2Fidea-to-build/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/winchxyz%2Fidea-to-build/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/winchxyz%2Fidea-to-build/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/winchxyz","download_url":"https://codeload.github.com/winchxyz/idea-to-build/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/winchxyz%2Fidea-to-build/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33930311,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-05T02:00:06.157Z","response_time":120,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agents","brainstorming","claude","claude-code","cowork-plugin","mcp","methodology","multi-agent","open-source","prompt-engineering"],"created_at":"2026-05-31T00:01:28.627Z","updated_at":"2026-06-05T05:00:49.284Z","avatar_url":"https://github.com/winchxyz.png","language":"Shell","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n\u003cimg src=\"assets/logo.png\" alt=\"idea-to-build — multi-agent brainstorming methodology\" width=\"800\"/\u003e\n\n# idea-to-build\n\n**A multi-agent brainstorming methodology that takes a raw idea to a plan you can build — for founders, builders, and creators.**\n\nTurn a raw idea into a plan you can actually build — in about 30 minutes. Claude sub-agents research it, argue with it, and won't let you skip the hard parts; when it's done, it hands the whole thing to Claude Code ready to build.\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)\n[![Claude](https://img.shields.io/badge/Powered%20by-Claude-D97757)](https://claude.com)\n[![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)\n[![Status](https://img.shields.io/badge/status-v0.4.0-orange.svg)](CHANGELOG.md)\n\n[Quick Start](#-quick-start) • [See It in Action](#-see-it-in-action) • [How It Works](#-how-it-works) • [Profiles](#-profiles) • [Why This](#-why-this-vs-alternatives) • [Contribute](CONTRIBUTING.md)\n\n\u003c/div\u003e\n\n---\n\n## The Problem\n\nBrainstorming with an LLM is shallow — and it dead-ends. You get an enthusiastic wall of text, and then nothing you can actually build from.\n\nAsk any LLM — Claude, ChatGPT, Gemini, Grok — to brainstorm and you hit the same failure modes:\n- ❌ Agreement, not pushback — it tells you why your idea is brilliant, not why it'll fail\n- ❌ Fabricated market sizes and made-up statistics\n- ❌ Generic SWOT lists that fit any business\n- ❌ \"5 great ideas!\" — none of which are actually different\n- ❌ No memory across sessions — you re-explain context every time\n- ❌ And even a *good* brainstorm ends in a chat you have to translate into a plan, then into a build, by hand\n\nThis tool fixes those — and closes the loop to a build. The full version runs on Claude (Cowork, Claude Code, or the Claude CLI); a degraded standalone prompt (`distributions/standalone-prompts/lite.md`) works in any other LLM.\n\n## The Solution\n\n**A structured 6-phase brainstorm that takes a raw idea all the way to something you can build.** Understand → Context → Generate → Deep Dive → Critique → Plan — each phase does the work most chats skip: real research with sources, genuinely different options, a ruthless critique that won't let you skip the premortem, and a plan with go/no-go gates.\n\nThen the part that makes it *idea-to-build*: when the brainstorm is done, **`/scaffold` turns the whole thing into a folder you open in Claude Code and build from** — the chosen approach, the paths you rejected, the risks to watch, the gates. Idea in, buildable plan out.\n\n\u003e \"It found a flaw in my pivot that I'd been missing for three weeks.\" — early tester\n\n---\n\n## 👀 See It in Action\n\nhttps://github.com/user-attachments/assets/7fa999e7-d4d8-4d67-b39e-ac2698c52ca6\n\nReal sessions, each linked in full — strongest first. Most put one prompt through a normal chat vs. the methodology, side by side; one runs a single idea through all six profiles; the first is a complete start-to-finish run that ends in a buildable plan:\n\n- [**Full run: medieval tycoon game**](examples/medieval-tycoon-fullrun.md) — the whole method, Phase 1 → 6 → `/scaffold`. A vague \"tycoon game\" comes out scoped, critiqued (GO with conditions), planned with a kill-switch, and handed off as a buildable folder. **Start here — it's the idea→build arc end to end.**\n- [**Hyperliquid wallet**](examples/hyperliquid-wallet-comparison.md) — the deepest comparison: the full method through a trace-verified isolated Phase-5 critique. Read this to see the rigor.\n- [**Personal health AI**](examples/health-system-comparison.md) — the harder case: plain Claude was already competent, and the methodology still added strategy (beachhead, regulatory risk, an architectural blocker).\n- [**Profiles in action**](examples/profiles-comparison.md) — the *same* idea run through six profiles, showing how each asks different questions. Proof the profiles aren't cosmetic.\n- [**Food delivery app**](examples/food-delivery-comparison.md) — the \"can it say no?\" test (trace-verified): the cold critic returns NO-GO on the head-on idea, then GO-with-conditions on the pivot.\n- [**AI support agent**](examples/ai-support-agent-comparison.md) — another honest NO-GO: the methodology pushes back and recommends *not* building the idea.\n- [**Stickman game**](examples/stickman-comparison.md) — the simplest contrast: plain Claude builds instantly, the methodology stops to scope. A quick way to see the difference.\n\n---\n\n## 🚀 Quick Start\n\nFour install paths — all available.\n\n### ✅ Option 1: Plugin via marketplace — Claude CLI (terminal)\nIn a terminal, run `claude`, then:\n```bash\n/plugin marketplace add winchxyz/idea-to-build\n/plugin install idea-to-build@idea-to-build\n```\nThen `/idea-to-build:start` to activate the coordinator. All ten commands are namespaced `/idea-to-build:*` (e.g. `/idea-to-build:critique`, `/idea-to-build:scaffold`). The `/plugin …` commands run in the Claude CLI; once installed, the `/idea-to-build:*` commands work in your CLI sessions. Verified working.\n\n### ✅ Option 2: Plugin via marketplace — Cowork / Claude Code (desktop app)\nIn the Claude desktop app (Cowork or Claude Code — one shared environment): **Customize → Personal plugins → `+` → Create plugin → Add marketplace**, enter `winchxyz/idea-to-build`, hit **Sync**, install the plugin, then run `/idea-to-build:start`. Once installed this way, the `/idea-to-build:*` commands work across both Cowork and Claude Code.\n\n\u003e **If Cowork's marketplace lags or won't update** — a known Cowork plugin-update bug where the *Update* button stays greyed out and re-adding reuses a stale cache — install it manually instead: download **`idea-to-build-plugin.zip`** from the [latest release](https://github.com/winchxyz/idea-to-build/releases/latest), then **Customize → Personal plugins → `+` → Create plugin → Upload plugin** and pick the zip. It lands under **My Uploads** and works identically.\n\n### ✅ Option 3: Clone the repo (open in the desktop app or the Claude CLI)\n```bash\ngit clone https://github.com/winchxyz/idea-to-build.git\n```\nOpen the cloned folder. The root [`CLAUDE.md`](CLAUDE.md) bootstraps the coordinator automatically (no `/start` needed); the repo's `.claude/skills/` provide the `/`-commands in Claude Code and the Claude CLI.\n\n### ✅ Option 4: Standalone prompt — any LLM\nCopy [`distributions/standalone-prompts/lite.md`](distributions/standalone-prompts/lite.md) into ChatGPT, Claude, Gemini, or any chat. Degraded quality (no sub-agent isolation, no cross-session memory, no `/scaffold`), but zero setup.\n\n---\n\n## 🧩 How It Works\n\nA coordinator guides you through six phases — and won't let you skip the ones that hurt. The critique phase is a ruthless, structured pass (forced premortem, steelman, inversion, a real verdict); the planning phase produces a gated plan with a kill-switch. When the plan is done, a scaffolder turns the whole brainstorm into a folder Claude Code can build from.\n\n```\n┌────────────────────────────────────────────────────────────────┐\n│                    Coordinator (you talk to)                   │\n│                                                                │\n│  Phase 1 ─ Understanding   \"What are you actually trying to    │\n│                             accomplish?\"                       │\n│  Phase 2 ─ Context         🔍 Research Agent (optional)        │\n│  Phase 3 ─ Generation      💡 Ideation Agent (optional)        │\n│  Phase 4 ─ Deep Dive       🔬 Deep-Dive Agent (optional)       │\n│  Phase 5 ─ Critique        ⚔️  In-context, adversarial        │\n│                             Sees: everything — stays harsh     │\n│                             Forced: premortem + steelman       │\n│  Phase 6 ─ Plan            📋 Planner Sub-Agent  ◄── isolated │\n│                             Sees: choice + critique            │\n│                             Forced: actionable steps + gates   │\n├────────────────────────────────────────────────────────────────┤\n│  /scaffold ───────────────▶ 📦 Scaffolder → buildable folder   │\n│                             (CLAUDE.md · README · DECISIONS ·   │\n│                              PLAN) → open in Claude Code, build │\n└────────────────────────────────────────────────────────────────┘\n```\n\n**Why this works:** Most \"brainstorms\" fail because the model skips the hard parts and then leaves you with a transcript. idea-to-build forces the hard parts — a structured premortem and steelman you can't skip, fact-checking with confidence labels, and a verdict that's allowed to be NO-GO — then hands the result to a build via `/scaffold`. The forced structure is the lever; the build handoff is the point.\n\n---\n\n## 📚 Profiles\n\nidea-to-build ships with **6 modes**: one general-purpose base plus five domain specializations.\n\n| Profile | When to use | Key tools |\n|---------|-------------|-----------|\n| 🎯 [General](profiles/general.md) | Any idea, any domain (calibrates to nature) | Full 6-phase framework |\n| 🚀 [Startup](profiles/startup.md) | Founder ideating a product/business | Unit economics, GTM, JTBD, Beachhead |\n| 🛠️ [Personal Project](profiles/personal-project.md) | A tool/app/game you're building for yourself | Build-vs-reuse, MVP, YAGNI, abandonment premortem |\n| 🔭 [Exploration](profiles/exploration.md) | Thinking a topic through; no build intended | First-principles, steelman, second-order effects |\n| ⚙️ [Tech Architecture](profiles/tech-architecture.md) | System design, stack choices | Trade-off matrices, RFC structure |\n| 🎬 [Content Strategy](profiles/content-strategy.md) | Channel, niche, monetization | Audience-fit, virality patterns, RPM |\n| 🗺️ [Product Roadmap](profiles/product-roadmap.md) | Feature prioritization, GTM | ICE/RICE, Kano, North Star Metric |\n| 🧭 [Personal Decisions](profiles/personal-decisions.md) | Career moves, life transitions | Reversibility, Expected Value, Inversion |\n\nA preset is chosen at the start of every session (or auto-classified from your description); it sets the lens **and** the flow shape — which phases run full, light, or skipped. Profiles override defaults inside each phase. Switch any time with `/profile startup` — a real slash command in Claude Code and the Claude CLI; in Cowork, just say it in plain language (\"switch to the startup profile\"). See six compared on one idea: [`profiles-comparison.md`](examples/profiles-comparison.md).\n\n---\n\n## 🆚 Why This vs Alternatives\n\n| | Raw ChatGPT/Claude | Awesome-prompts repo | LangChain/CrewAI build | **idea-to-build** |\n|---|---|---|---|---|\n| Setup time | 0 sec | 30 sec | 30+ min | ✅ 60 sec |\n| Forced critique | ❌ | ⚠️ Often skipped | ✅ if coded | ✅ Hard-gated |\n| Fact-checking discipline | ❌ Hallucinates | ❌ No enforcement | ⚠️ DIY | ✅ Tier 1/2/3 protocol |\n| Cross-session memory | ❌ | ❌ | ✅ if coded | ✅ Context files |\n| Isolated planner | ❌ | ❌ | ✅ if coded | ✅ Fresh-context sub-agent |\n| Domain profiles | ❌ One-size | ❌ | ⚠️ DIY | ✅ 8 ready + auto-preset |\n| Ends in a buildable plan | ❌ Just a chat | ❌ | ⚠️ DIY | ✅ `/scaffold` → folder Claude Code builds |\n| Cost per brainstorm | Free–$1 | Free | $$ (LLM API) | Free (your Claude subscription) |\n\nMore questions — *is it really multi-agent? could a prompt do this? does it actually say no?* — see the [**FAQ**](docs/FAQ.md).\n\n---\n\n## 🎯 Core Principles\n\n1. **Skeptical by default.** Every claim — yours, the user's, the source's — is a hypothesis to be tested. Accuracy over confidence, clarity over speed, evidence over assumption. When uncertain, the tool tells you exactly what would verify the claim.\n2. **Phase-explicit communication.** Coordinator announces the current phase in every message. No silent transitions.\n3. **Factual rigor (Tier 1/2/3).** Made-up numbers are a sin. Each material claim gets a ✅ / ⚠️ / 🔍 confidence label.\n4. **Calibrated recommendations.** Every recommendation includes a confidence percentage, what would raise/lower it, and an alternative if confidence drops. No false certainty.\n5. **Forced critique.** Phase 5 cannot be skipped. Premortem + What-Needs-to-Be-True are mandatory.\n6. **Adversarial critique.** A dedicated critic pass forced to premortem, steelman, and invert — built to find the flaw you're too attached to see, and to return an honest GO / NO-GO.\n7. **Memory as a log, not state.** Decisions and rejected options are appended, never overwritten. History matters.\n8. **Ends in a build, not just a brainstorm.** The point isn't a tidy writeup — it's a scoped, critiqued plan you can act on. `/scaffold` turns the finished brainstorm into a folder Claude Code builds from.\n\nFull methodology: [`docs/METHODOLOGY.md`](docs/METHODOLOGY.md)\n\n---\n\n## 🏗️ Architecture\n\n```\nidea-to-build/\n├── CLAUDE.md                 # Entry point — auto-activates the coordinator\n├── .claude/skills/           # Slash commands for Claude Code and the Claude CLI (/profile, /critique…)\n├── .claude-plugin/           # Marketplace manifest (Add marketplace / /plugin install)\n├── core/\n│   ├── CLAUDE.md             # Full coordinator specification\n│   ├── agents/               # 6 agents: research, ideation, deep-dive, critic, planner, scaffolder\n│   ├── skills/               # Recommendation + confidence module\n│   └── templates/            # Project context file template\n├── profiles/                 # 8 domain profiles (general + 7 specialized)\n├── distributions/\n│   ├── claude-code-plugin/   # Installable plugin — 10 commands + 6 agents (Cowork / Claude CLI)\n│   └── standalone-prompts/   # Lite version for any LLM\n├── docs/\n│   ├── ARCHITECTURE.md       # How sub-agents are orchestrated\n│   ├── METHODOLOGY.md        # Core principles in depth\n│   ├── PHASES.md             # 6 phases explained\n│   ├── PROFILES.md           # Profile authoring guide\n│   └── FAQ.md                # Common questions, answered honestly\n└── examples/                 # Template, guide + real comparison transcripts\n```\n\n\u003e The `examples/` directory ships with a template, an authoring guide, and seven real transcripts — five run the same prompt as plain Claude vs. through the methodology: a stickman game, a personal-health-AI startup, a Hyperliquid wallet (full methodology through a trace-verified Phase 5 critique → NO-GO), an AI support agent, and a food-delivery app (where the Phase 5 critique returns NO-GO on the head-on idea, then GO-with-conditions on the pivot) — plus a profiles comparison (one idea, six profiles) and a full start-to-finish run that ends in a buildable plan. A broader library lands in v0.2. See [`CHANGELOG.md`](CHANGELOG.md).\n\nFull architecture: [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md)\n\n---\n\n## 🤝 Contributing\n\nWe welcome:\n- 🆕 New profiles for specific domains (legal, healthcare, education, climate, etc.)\n- 🛠️ Improvements to sub-agent prompts\n- 📖 Real-world brainstorm transcripts for `examples/`\n- 🌍 Translations (currently English-first; methodology language-agnostic)\n- 🐛 Bug reports and edge cases\n\nSee [`CONTRIBUTING.md`](CONTRIBUTING.md).\n\n---\n\n## 📜 License\n\nMIT — see [`LICENSE`](LICENSE). Use it, fork it, ship it commercially. Just don't sue us if your brainstorm convinces you to do something dumb.\n\n---\n\n## ⭐ Star History\n\nIf `idea-to-build` helped you avoid a bad decision (or commit to a good one), a star makes our day.\n\n[![Star History Chart](https://api.star-history.com/svg?repos=winchxyz/idea-to-build\u0026type=Date)](https://star-history.com/#winchxyz/idea-to-build\u0026Date)\n\n---\n\n\u003cdiv align=\"center\"\u003e\nBuilt with 🧠 by people who got tired of LLMs agreeing with them.\n\u003c/div\u003e\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwinchxyz%2Fidea-to-build","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwinchxyz%2Fidea-to-build","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwinchxyz%2Fidea-to-build/lists"}