Launchpad

An automated feature launchpad: every feature starts from structured intent and comes out the other end as a developer-reviewable pull request.

intent in → Figma artifact out → Claude/API reasoning → scaffolded code → validation evidence → PR

This is not intended to replace designers, architects, or developers. It's meant to produce the first serious draft of a feature — visual design, code scaffold, validation notes, and an open PR — so a human can review and take it from there.

Status: partial implementation. Stages 1-3 of the pipeline (normalize intent → UX flow → screen map) are implemented in feature_launchpad/; Stages 4+ (Figma generation, code scaffold, validation, PR) don't exist yet because intent.md itself cuts off mid-way through "Stage 3 — Create Screen Map". See CLAUDE.md for a fuller breakdown of what's defined so far and what's still missing.

Running it today

export ANTHROPIC_API_KEY=sk-ant-...
python3 -m feature_launchpad ./intents/customer-billing.intent.md

This runs Stages 1-3 and writes generated/feature-launchpad/customer-billing/: 01-normalized-intent.md, 02-ux-flow.md, 03-screen-map.json, and a launchpad-run-log.json. If ANTHROPIC_API_KEY isn't set, the workflow stops immediately with a clear message and writes nothing, per intent.md Section 5.

Run the test suite (standard library unittest, no extra dependencies):

python3 -m unittest discover -s tests -v

Target experience (not yet fully implemented)

/feature-launchpad ./intents/customer-billing.intent.md

produces a feature/customer-billing-dashboard branch containing generated design/planning artifacts (UX flow, screen map, Figma payload, implementation plan, validation report, PR summary) alongside the actual code and test changes. Today's CLI produces the first three generated artifacts only, on the current branch — branch/PR automation is part of the unimplemented Stages 4+.

Example: an intent taken all the way to a working app

Since Stages 4+ don't exist yet, blackjack-game/ is a hand-written (not pipeline-generated) implementation of intents/blackjack-game.intent.md, built from that intent's Stage 1-3 output. It's a standalone client-only Vite + React app — see blackjack-game/README.md to run it. This shows what a Stage 4+ code-scaffolding step would need to produce, without claiming the pipeline actually does this yet.

Core principle

• The intent file is the source of truth.
• Figma is the visual execution layer.
• Claude/Anthropic reasoning is the design and implementation intelligence layer.
• The repository is the delivery layer.
• The pull request is the handoff artifact — a human always reviews before merge.

Constraints

Version one is meant to be simple, local, and dependency-light: no orchestration frameworks (LangChain, CrewAI, AutoGen, etc.), no database, no queue, no server, no web UI, no Docker, no cloud deployment. Just direct HTTP calls to the Anthropic and Figma APIs, using each language's standard library wherever possible. Every run happens on a new feature branch, and existing source is never overwritten without a backup or diff.

The one exception is an optional, read-only status dashboard (python3 -m feature_launchpad.dashboard, see USER_MANUAL.md) — a static HTML snapshot, not a running server, that shows environment health and current tool capabilities. The pipeline itself still needs no server and no web UI.

Reading more

• USER_MANUAL.md — step-by-step guide to running the CLI today: setup, writing an intent file, reading the output, and troubleshooting.
• intent.md — the original (partial) spec: mission, required env vars, input intent file format, output artifacts, and workflow stages.
• CLAUDE.md — guidance for Claude Code sessions working in this repo, including where the spec is ambiguous or incomplete.
• feature_launchpad/ — the Stage 1-3 implementation (stdlib-only Python; see module docstrings for how each stage maps to intent.md Section 8).
• CONTRIBUTING.md — how to contribute right now (mostly: closing spec gaps).
• CHANGELOG.md — notable changes, following Keep a Changelog / SemVer.
• SECURITY.md — how to report a vulnerability; secret-handling expectations.
• CODE_OF_CONDUCT.md — community standards for this project.

Project status

Implementation	Stages 1-3 (feature_launchpad/); Stages 4+ don't exist yet
Spec completeness	Partial — truncated mid "Stage 3" (see CLAUDE.md)
Version	0.2.0 (see VERSION and CHANGELOG.md)
License	Apache 2.0

Contributing

See CONTRIBUTING.md. High-value contributions right now: closing gaps in intent.md (it cuts off mid-schema in Stage 3), or implementing Stage 4 (Figma generation) once that gap is resolved.