Open source tools for software engineering teams using AI.

Decades of computer science and software engineering research — structured product discovery, test-driven development, formal methods, and behavior-preserving transformations — have always worked. They were just challenging to apply consistently under pressure. We believe AI changes that equation. We build open source CLI tools, Claude Code plugins, and MCP servers to find out.

Our design standards and project guide live in punt-kit. Our tenets document where we think agentic coding is heading.

• PR/FAQ Claude Code — Structured product discovery using Amazon's Working Backwards process. Eight specialized agents, simulated review meetings, compiled PDF output. Grounding for what to build.
• Use Cases — Guided use-case specification using Jacobson-Cockburn Use-Case Foundation v1.1. In development.
• Z Spec Claude Code — Formal Z specifications for stateful systems. Extract specs from existing code (code2model) or generate code from specs (model2code). Type-check with fuzz, animate and model-check with ProB. Grounding for how to build it.
• Refactory — Deterministic, behavior-preserving refactoring. The AI decides what needs restructuring, Refactory executes on a parsed program model with formally defined preconditions. Built on Opdyke's original definition. In development.
• ReasonTrace — Session recordings as engineering assets. Captures every AI coding session as a dual-stream recording (terminal for humans, structured events for agents), extracts reasoning decisions, and runs automated review agents against project standards and formal specs. Agents reviewing agents. In development.

• Quarry CLI Claude Code Claude Desktop macOS — Local semantic search across your documents. 30+ formats, AST-level code parsing, offline. Works as a standalone CLI, an MCP server, or through the menu bar app.
• Beadle MCP Go — Email communication over Proton Bridge with a four-level PGP trust model. Written in Go.
• Biff CLI Claude Code — Team communication resurrecting BSD Unix vocabulary (who, finger, write, plan) over a NATS relay. Humans and agents show up side by side. Serious coordination, retro charm.
• Vox CLI Claude Code — Text-to-speech engine with multi-provider support (ElevenLabs, OpenAI, AWS Polly). Spoken notifications, chimes, and arbitrary speech synthesis. Opt-in only.
• Persona — Character, voice, and teaching philosophy for every domain tool. Grounded in Mollick & Mollick's seven pedagogical roles. The pattern already works in langlearn-tts's 28 named instructors. In development.
• Lux CLI Claude Code — A visual output surface for AI agents. ImGui display server connected by Unix socket IPC — agents send JSON element trees, the display renders them at 60fps. 22 element kinds, interactive controls, incremental updates, render functions with consent.
• Tally — Token consumption, model usage, and spend tracking across sessions and projects. In development.
• punt-kit CLI Python — Org standards, shared tooling, and project scaffolding. The punt CLI.

Real projects where we discover what works and what doesn't.

• Koch Trainer iOS — Morse code trainer using the Koch method. Built from a formal Z Spec specification — the design-to-code workflow in action.
• langlearn-tts Claude Desktop — An AI language tutor that speaks to you. 28 tutor personas across 7 languages and 4 levels. Built on Vox.
• LangLearn — The complete platform wiring langlearn-tts, langlearn-anki, and langlearn-imagegen into one pipeline. In development.
• Dungeon Claude Code — Text adventure where Claude is the game master. No code runs, only prompts. Built to experiment with UX concepts for Claude Code experiences.

All CLIs and all Claude Code plugins in a single command:

curl -fsSL https://raw.githubusercontent.com/punt-labs/punt-kit/f61167d/install-all.sh | sh

All Python packages are published under the punt- prefix on PyPI: