The Thought Layer Kit
The same rigor that runs in the web app also runs inside your own coding agent, as portable Agent Skills, a Pi package, and a CLI. Validate the idea, grill it into a buildable spec, then build and deploy it. Your own key, nothing phones home.
AI made building cheap. It did not make knowing what to build cheap, and it did not make a confident, defensible plan cheap. The Thought Layer Kit puts that rigor where building happens, then carries it all the way to a live thing you own. It is open source (MIT) and bring your own key by design. The point is to help people build real things instead of confident slop, not to sell a platform.
The loop it closes
Validate the idea, grill it into a buildable spec, build it, deploy it. One agent. Your own key. Nothing phones home.
The loop, in order
- Validate the idea. What it is, your domain knowledge, demand, the market, the 30-second pitch. A panel judges each at the idea's altitude, so a one-line idea is never audited for implementation details.
- Make the model real. Time, costs, scale, pricing, the business model and its numbers, acquisition, relationships, support.
- Design it (last). Draft the PRD with a first-cut glossary and testable requirements, then grill that draft against the domain until it is build-ready.
- Build it. The agent builds the hardened PRD static-first into a predictable publish directory, honoring the glossary, requirement traceability, and the SEO layer.
- Deploy it. Take the build live to a URL you own, with no lock-in.
Install it anywhere
The skills are plain Markdown and work in any agent that reads the Agent Skills format. The Pi package adds deterministic tools and slash commands. The CLI runs in any shell.
Pi
pi install npm:@hobocode/thought-layer
Claude Code (or any agent that reads Agent Skills)
git clone https://github.com/hobocode-ofc/thought-layer-kit
cp -r thought-layer-kit/skills/* ~/.claude/skills/
Any shell agent (the CLI)
npx -y @hobocode/thought-layer tl read
What is inside
The rigor, as 12 Agent Skills
| Skill | What it does |
|---|---|
| thought-layer-framework | The backbone. Walks the full staged framework in order and only reaches design at the end. |
| thought-layer-panel | Pressure tests one stage with an adversarial panel at that stage's altitude. Confidence score, grade, up to three fixes. |
| thought-layer-prd | Drafts the complete PRD from the validated idea and business model, with a glossary and testable requirements. |
| thought-layer-grill | Grills the draft PRD against the domain, one question at a time, and hardens it inline into a build-ready spec. |
| thought-layer-naming | Proposes names grounded in the business, each with a rationale, a style, and a domain-ready slug. |
| thought-layer-build | Builds the hardened PRD into a static-first, deploy-ready artifact, verified to run. |
| thought-layer-deploy | Takes the build live to a URL you own, with no lock-in. |
| thought-layer-speedrun | A fast, unranked path to a build-ready spec: no panel, no score, no gate. |
| thought-layer-strategy | Optional deep dive: positioning, the moat mechanism on an honest clock, beachhead to expansion, the incumbent response. |
| thought-layer-brand | Optional deep dive: the emotional promise, personality and archetype, voice, a name, and a visual brief. |
| thought-layer-market-research | Optional deep dive: the buyer and trigger, two sizings, demand evidence, competitors, willingness to pay, the wedge, why now. |
| thought-layer-business-model | Optional deep dive: the money flow, per-unit economics with every number sourced, pricing, and base, bull, and bear scenarios. |
Six deterministic tools (Pi)
These run as real code, not the model, so the scoring and the math are exact and never re-derived.
| Tool | What it does |
|---|---|
| tl_score | Aggregate confidence values into a stoplight status and a letter grade, using the exact bands. |
| tl_domains | Check domain availability across common TLDs (bring your own RapidAPI key, or a search link). |
| tl_project | Run the monthly business projection: break-even, year-one revenue and net, max drawdown, ending MRR. |
| tl_state | Read and write the portable progress file shared losslessly with the web app. |
| tl_scaffold | Generate a branded, SEO-complete static landing site as an instantly deployable floor. |
| deploy | Take the built site live to a user-owned Netlify URL. |
Eight slash commands and a CLI
Slash commands map to the stages: /tl runs the whole flow, with /tl-speedrun, /tl-panel, /tl-prd, /tl-grill, /tl-naming, /tl-build, and /tl-deploy for each step. The tl CLI gives any shell agent the same state file, scaffold, and deploy operations. See the Pi guide and the Claude Code guide.
It speaks the web app's language
The kit reads and writes the same .thought-layer/state.json progress file the web app exports. A non-technical founder in the browser and a technical co-founder in an agent can pass one lossless file back and forth, resuming and collaborating across both surfaces.
