Overview

pstack is an advanced, opinionated agent skill stack, meticulously ported to support a wide range of AI harnesses including Claude Code, Codex, Pi, Gemini, and Prime Agent. Building on Lauren Tan's original Cursor skill stack, this port enhances agent workflows by translating Cursor primitives into robust, harness-specific implementations. It's designed for rigorous execution, ensuring your code remains concise, simple, and thoroughly verified.

Key Features

• Multi-Harness Support: Seamlessly integrates with Claude Code, Codex, Pi, Gemini, and Prime Agent, offering a consistent, high-quality agent experience across diverse AI development environments.
• Opinionated Workflows: Leverages pre-defined, effective poteto-mode workflows tailored for tasks like bug fixing, planning, refactoring, and more, streamlining the development process.
• Rigorous Verification: Employs strategies such as how, why, and interrogate to thoroughly investigate issues, delegate fixes, and verify solutions with passing evidence.
• Modular Design: Carries named policy forks and is designed for easy extension and maintenance, with detailed documentation for skills, commands, and runtime setup.
• Local Execution: Operates with a strong focus on privacy, featuring no server or telemetry. All scripts run locally, and PR tools utilize your GitHub CLI login.

Typical Use Cases

• Automated Bug Fixing: Provide poteto-mode with a bug description, and it will reproduce the failure, investigate the root cause using advanced reasoning tools, delegate the fix, and present you with the verified solution and evidence.
• Complex Development Tasks: Utilize specialized playbooks for planning new features, refactoring codebases, addressing performance bottlenecks, prototyping, and managing Pull Requests.
• Proactive Code Verification: Ensure the integrity and conciseness of your code through automated workflows that emphasize simplicity and verification at each step.
• Customizable Agent Behavior: Configure model defaults and reasoning efforts for different roles using setup-pstack to fine-tune agent performance for specific tasks and projects.