michael-denyer/pstack-claude

Claude Code, Codex, Copilot, Pi, OpenCode, Muse, Gemini, and Prime Agent versions of Poteto's pstack. Rigorous agent workflows with Cursor primitives translated for other harnesses.

What it solves

It improves the outcomes of AI coding agents by providing an opinionated "skill stack" of workflows. It prevents agents from producing sloppy code and ensures that fixes are verified through a structured process of reproduction, investigation, and testing.

How it works

The project ports a set of specialized skills and playbooks to various agent harnesses (such as Claude Code, GitHub Copilot, Codex, and Pi). Users can invoke poteto-mode to trigger a specific workflow based on their goal. For example, when fixing a bug, the system reproduces the failure, uses investigation tools (how and why), delegates the fix (potentially involving an architect role if boundaries are crossed), and reruns tests to provide passing evidence.

Who it’s for

Developers using AI agent harnesses like Claude Code, GitHub Copilot, Codex, or Pi who want more reliable, concise, and verified code generation.

Highlights

  • Multi-Harness Support: Compatible with Claude Code, Codex, Pi, and GitHub Copilot.
  • Poteto-mode: A routing mechanism that invokes the correct workflow (playbook) for tasks like bug fixing, refactoring, or prototyping.
  • Role-Based Configuration: Allows setting specific models and reasoning effort levels for different roles (e.g., arena runners, fable).
  • Verified Workflows: Integrates reproduction and verification steps into the agent's process to ensure fixes actually work.

Related

  • Project
    ericlitman/open-pstack

    A plugin for Claude Code and Codex that implements a rigorous engineering framework for coding agents, emphasizing verification, multi-model review, and structured workflows.

  • Project
    feiskyer/claude-code-settings

    A set of Claude Code plugins (skills & sub‑agents) that add brainstorming, deep research, GitHub issue/PR automation, image generation, translation, and more, with ready‑made config files for many LLM providers.

  • Project
    umputun/cc-thingz

    A marketplace of Claude Code plugins that add brainstorming, PR review, planning, release automation, thinking aids, skill‑evaluation hooks, and workflow helpers. Install via `/plugin marketplace add umputun/cc-thingz` and pick the plugins you need.

  • Project
    backnotprop/pstack

    A collection of agent-based engineering skills and playbooks designed to help AI agents write high-quality, verifiable code by automating rigorous development workflows.

  • Project
    Piebald-AI/claude-code-lsps

    A marketplace of LSP server definitions that lets Claude Code (Anthropic’s AI coding assistant) provide IDE‑like features—go‑to‑definition, hover, find‑references, etc.—for dozens of languages. Install Claude Code ≥ 2.1.50, run `npx tweakcc --apply` to patch it, add the marketplace via `/plugin marketplace add Piebald‑AI/claude‑code‑lsps`, then install the underlying language‑server binaries (rust‑analyzer, pyright, clangd, etc.) as instructed. The repo includes validation scripts to keep the marketplace JSON in sync with the `.lsp.json` configs.