backnotprop/pstack

What it solves

It addresses the problem of "slop" code—low-quality, AI-generated code that lacks rigor. It provides a structured, engineering-focused framework to help developers use AI agents to ship high-quality, verifiable software rather than just maximizing lines of code.

How it works

It functions as a collection of "Agent Skills" that can be integrated into various AI coding environments (like Claude Code, Cursor, or Pi). The core component is /poteto-mode, which acts as a central orchestrator. When you provide a task, it selects a specific "playbook" (such as bug fixing, performance profiling, or architectural design), breaks the task into a todo list, and automatically triggers specialized sub-skills to handle individual steps, verification, and documentation.

Who it’s for

It is designed for software engineers who want to use AI to perform rigorous, multi-step engineering tasks—such as refactoring, formal verification, or complex bug hunting—without sacrificing code quality or reliability.

Highlights

• Rigorous Engineering Playbooks: Includes 23 specialized playbooks for tasks like hillclimbing metrics, runtime forensics, and autonomous PR management.
• Multi-Model Orchestration: Designed to leverage the unique strengths of different frontier models (e.g., using one model for architectural judgment and another for code implementation).
• Verifiable Results: Emphasizes "fearless parallelism" by ensuring agents perform verification steps, such as writing failing tests first (TDD) or using formal verification for complex state machines.
• Extensible Framework: Works as a set of standard Agent Skills, allowing it to be used across different AI agent harnesses rather than being locked to a single platform.