The Kit in Pi

Pi is the flagship host for the kit. Install the package, run the framework with /tl, and let the deterministic tools keep the scoring and the math exact while the agent does the thinking.

What Pi is

Pi (@earendil-works/pi-coding-agent, by Mario Zechner) is a minimal, MIT, bring your own key coding agent. You extend it with skills, prompt templates, and TypeScript tools, distributed over plain npm or git. Its ethos, a minimal core extended by packages with no SaaS and no lock-in, mirrors the kit's, which is why it is the flagship host.

Install

pi install npm:@hobocode/thought-layer

# or, to track the latest from GitHub during development:
pi install git:github.com/hobocode-ofc/thought-layer-kit

Installing the package lights up the skills, the /tl commands, and the tl_score, tl_domains, tl_project, tl_state, tl_scaffold, and deploy tools. You can also invoke a skill directly, for example /skill:thought-layer-panel.

Run the framework with /tl

/tl walks the whole framework in order and does not skip ahead. It asks for your idea, then moves through validate, model, and design last. It advances exactly one stage per turn and stops for you.

One stage at a time

Each stage is judged by an adversarial panel (a red team, a domain expert, and a skeptical investor) at that stage's altitude. You get a confidence score, a letter grade, and at most three fixes that belong to this stage. A stage clears at 0.85 confidence, or when you set it aside, in which case the open items carry forward as to-dos.

The deterministic tools

The kit registers real tools so the agent never re-derives the math. This is where the consistency lives.

Tool	What it does
tl_score	Aggregate one or more confidence values into a stoplight status and a letter grade.
tl_domains	Check domain availability across common TLDs, reading a RapidAPI key from the environment.
tl_project	Run the monthly projection: break-even, year-one revenue and net, max cash drawdown, ending MRR.
tl_state	Read and update the portable progress file (the same file the web app uses).
tl_scaffold	Write a branded, SEO-complete static site as an instantly deployable floor.
deploy	Take the publish directory live to a user-owned Netlify URL.
tl_sync	Store and version your session files in your own private GitHub repo over git. BYOK, no token paste.

Build and deploy, in the same agent

After the grill hardens the PRD, the build and deploy steps close the loop without leaving Pi.

• Build with /tl-build: the agent builds the spec static-first into a predictable dist/, honoring the glossary, requirement traceability, the out-of-scope list, mobile and desktop, brand, and the full SEO layer. A three question test gates any backend and flags it loudly.
• Scaffold with tl_scaffold for the fastest deployable floor: a branded, SEO-complete landing site, with no model call.
• Deploy with the deploy tool: with a Netlify token it deploys into your own account via the file-digest API (owned immediately); with no token it uses the Netlify CLI, creating a site in your account when logged in, or an anonymous, claimable URL when logged out.

Sync and collaborate

Keep your sessions in your own private GitHub repo so you can move between machines and collaborate with others. The kit syncs over git, and no token is pasted into the CLI (it uses your gh login or your git credential helper). Each session is one file in the repo.

tl sync init --repo your-name/tl-sessions   # gh creates the private repo if needed
tl sync save --name photobooth              # snapshot the session, commit, push
tl sync list                                # the sessions in the repo
tl sync open --name photobooth              # pull and resume a session

# or with no install, via npx:
npx -y @hobocode/thought-layer tl sync save --name photobooth

Reference

Slash commands

Command	Purpose
/tl	Run the full framework, one stage at a time.
/tl-speedrun	The fast, unranked path to a build-ready spec.
/tl-panel	Pressure test one stage answer with the panel.
/tl-prd	Draft the PRD from the validated idea and business model.
/tl-grill	Grill the draft PRD and harden it inline.
/tl-naming	Propose names with rationale and domain-ready slugs.
/tl-build	Build the hardened PRD into a deploy-ready artifact.
/tl-deploy	Take the build live to a URL you own.

Environment variables

Variable	Used for
THOUGHT_LAYER_STATE	The default state file path, so you can keep several ideas as separate files.
THOUGHT_LAYER_DOMAIN_KEY (or RAPIDAPI_KEY)	Your RapidAPI key for live domain checks. With no key, naming links to a search instead.
NETLIFY_AUTH_TOKEN	Your Netlify token for the deploy tool. Read from the environment only, never a parameter.