pstack-claude: Agent workflow that handles everything from reproduction to verification given only a goal

When a developer inputs a goal or bug in a single line, it automatically performs the entire process from code reproduction, root cause investigation, and fixing to verification. Beyond simple code generation, it provides a rigorous workflow that re-runs failed test cases to confirm resolution.

It is designed as a plugin that works across various AI agent harnesses such as Claude Code, Codex, and Pi. When complex modifications crossing function boundaries are required, it automatically invokes a sub-agent acting as an architect to maintain structural consistency.

This project is a port of the original pstack Cursor skill to other environments, adhering to the principle of local-based execution without servers or telemetry. For verifying concurrency bugs or invariants that are difficult to catch with tests, it can integrate with separate plugins supporting TLA+ and Lean.