Cursor's pstack, ported to the harness you actually use

What it does The front door is poteto-mode: tell it your goal and it picks the right playbook from a stack of skills originally built as a Cursor plugin and translated here for Claude Code, Codex, Pi, OpenCode, Gemini CLI, and Prime Agent. For a bug, it reproduces the failure, investigates with how and why, delegates the fix, then reruns the failing case — you get the fix plus the failing and passing evidence. Further playbooks cover planning, features, refactoring, performance, investigations, prototypes, PR maintenance, shipping, and longer projects.

The interesting bit The engineering is translation, not invention: Cursor’s primitives — skills, hooks, slash commands — remapped onto other runtimes, with a routing hook that intercepts requests on Claude Code and Codex so the playbook choice happens before the agent starts improvising. There’s also boundary discipline: a fix that crosses a function boundary pulls in architect before implementation, not after.

Key highlights

• poteto-mode routes a plain-language goal to the right workflow — plan, bugs, features, refactor, and more.
• Bug fixes come with receipts: the failing case before, the passing case after.
• A supporting cast — how, why, unslop, interrogate, tests, measurements, arena, swarm — handles investigation, review, and verification.
• Skills-only installs extend it to any harness, not just the six with dedicated support.
• Model defaults and automatic routing are configurable via setup-pstack.

Caveats

• As a faithful port, it tracks Lauren Tan’s upstream pstack; port scope and maintenance details live in docs/reference.md rather than the README.
• Automatic routing is on by default via an installed hook, and Codex makes you explicitly trust that hook before it runs — welcome discipline or an extra nag, depending on your temperament.
• “Improves agent outcomes” is asserted, not benchmarked; the rigor lives in the workflow design, not in published numbers.

Verdict Worth a look if you run Claude Code, Codex, or Gemini CLI and want your agent working from playbooks instead of vibes. If you’d rather author your own workflows, this is someone else’s opinions — faithfully translated, but someone else’s.