Search skills/

pstack-claude Claude skills

A port of Lauren Tan's pstack workflow skills to Claude Code, Codex, Copilot and others, with a poteto-mode router and 67 skills.

michael-denyer/pstack-claude 67 skills Code review & qualityPlanning, TDD & debugging
Our verdict Recommended with caveats

A capable workflow stack, but it is a port of someone else's work, skills depend on each other, and the full set is heavy.

3/5 Depth expertise shipped Note Security Review before installing 1d Maintenance Updated today 3.6K Adoption stars OK License MIT, commercial OK

What pstack-claude does

pstack is an opinionated skill stack from Lauren Tan, originally written for Cursor. This repository ports it to Claude Code, Codex, GitHub Copilot, Pi, OpenCode and other agents, translating the Cursor-specific tool names ahead of time. The entry point is poteto-mode: you state a goal and it routes to the right workflow. Skills include architect (sketch types and signatures before code), arena (several model perspectives on a design), blast-radius, babysit, fix-ci, fix-merge-conflicts, get-pr-comments, how and why (explain a subsystem and its history), deslop, no-comments, and a set of principle-* skills that act as rules.

There are 67 skills averaging about 4 KB each, plus TypeScript and Node scripts under poteto-mode for plan checking, resuming runs, worktree audits and watching pull requests. Those scripts need bun. A Claude Code SessionStart hook routes tasks to poteto-mode. The README says there is no server or telemetry, and that pull-request tools use your GitHub CLI login. Some skills call others, so installing a single one can leave it missing its dependencies. It tracks upstream and adds named policy forks.

Good for

  • Engineers who like a strict, review-heavy agent workflow
  • Teams already using Cursor's pstack who need it on other agents
  • PR babysitting and CI-fixing loops with a GitHub CLI login

Not for

  • Users who want a small, minimal skill set
  • Setups without bun or the GitHub CLI

What's inside depth 3/5

67 skills averaging 4 KB, a few with references, plus TypeScript orchestration and PR-watching scripts with tests, but no skill evals.

Skills
67
Reference docs
42
Scripts
84
Tests / evals
388
Templates
0
Avg SKILL.md
4.1 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • bun for the poteto-mode scripts
  • GitHub CLI login for PR tools
  • Claude Code 2.1.292 or later for the plugin route
  • Telemetry: none found; the README says pstack has no server or telemetry.

Security review

Review before installing

A multi-runtime port of the pstack skill set; its code is bun/node tooling that watches and inspects GitHub pull requests through your gh login, keeps an orchestration store on disk, and audits git worktrees (including reading local agent transcripts to date the last chat). The README states there is no server or telemetry. Playbooks cover merging and force-with-lease pushes but require your explicit landing authority. Review the plugin's default-on session hook (opt-out available), which was not in the files we could read.

Worth knowing

  • Installs packages or toolsOn first run the scripts execute 'bun install --frozen-lockfile --production' to fetch the commander package from the npm registry.plugins/pstack/skills/poteto-mode/scripts/bootstrap.ts:40
  • Relies on a CLI or package we could not reviewPR tools call the gh CLI (graphql, pr list/view/checks) and git using your existing GitHub login; bun is required to run the scripts.plugins/pstack/skills/poteto-mode/scripts/watch-pr/github.ts:521
  • Makes network requestsReads PR state and can run disablePullRequestAutoMerge/dequeuePullRequest GraphQL mutations through gh, only via the explicit ship-pr cancel-pending command; it never merges.plugins/pstack/skills/poteto-mode/scripts/watch-pr/shipping.ts:130
  • Makes network requestsRuns git ls-remote/fetch against origin and gh pr list; it also reads local Claude, Codex, Pi and Copilot session transcripts to find the last chat per worktree, read-only and local.plugins/pstack/skills/poteto-mode/scripts/worktree-audit.mjs:238
  • Changes files outside your projectThe orchestrate playbook tells the agent to create a state store under ~/.claude/orchestrate/<project>/ outside the repository; Copilot/Codex setup also writes a model sheet in the runtime's home directory.plugins/pstack/skills/poteto-mode/playbooks/orchestrate.md:25
  • installs_hooksThe plugin documents a SessionStart routing hook, on by default (disable with 'session hook: off'), and for Copilot a PreToolUse hook that auto-approves registered pstack scripts and denies mismatched model choices; Codex and Muse require approval, Claude Code and Copilot do not. The hook files are not in our snapshot.plugins/pstack/skills/poteto-mode/references/copilot-tools.md:58

What we checked

The repository (782 files) on Oct 29, 2026. Not a full review: 660 of 782 files were not downloaded (large repository; we reviewed the skill folders); we read 21 of 32 code files; it relies on code outside the repository: plugin hook scripts (session-start.sh/.ps1, Copilot PreToolUse hook, check-sheet.sh, log.sh) and plugin manifests, not in the files we have; npm package commander 14.0.0 (installed by bun at first run); bun runtime and GitHub CLI (gh) used by the scripts.

NVIDIA SkillSpector static scan: 31 pattern hits (Agent Snooping 14, Rogue Agent 4, Excessive Agency 4, Tool Misuse 3, Data Exfiltration 2, Supply Chain 2). SkillSpector hits are documentation mentions: git push --force-with-lease and reset --hard in playbooks (guarded procedures), ~/.codex and ~/.claude paths in runtime notes, 'nohup' and skill-path references. The scripts contain no direct HTTP calls, no credential reads (the url.password hits only reject credentialed URLs) and no obfuscation.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT — commercial use allowed; upstream pstack and imported Cursor skills carry their own copyright notices.

Can it run unattended?

Yes. Skills such as babysit, fix-ci and the PR watcher are built for long-running loops, with a plan checker and resume script.

Alternatives to pstack-claude

Superpowers

obra/superpowers

A 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.

Recommended Check first Depth 3/5

Matt Pocock's Skills

mattpocock/skills

Small, composable engineering skills: grill-me interviews, shared glossary, TDD, triage, prototyping, code review, handoff and a flow router.

Recommended Check first Depth 2/5

Agent Skills (Addy Osmani)

addyosmani/agent-skills

25 engineering-lifecycle skills and 9 slash commands (spec, plan, build, test, review, ship) that make coding agents follow senior-engineer workflows.

Recommended Check first Depth 3/5

FAQ

What is pstack-claude?

It is a port of Lauren Tan's pstack skill stack to Claude Code, Codex, Copilot, Pi and other agents. It adds a poteto-mode router and 67 skills for design, review, CI and PR work.

How do I install pstack-claude?

In Claude Code, run /plugin marketplace add michael-denyer/pstack-claude then /plugin install pstack@pstack-claude. Codex, Copilot and Pi have their own commands in the README.

pstack-claude vs backnotprop/pstack: which one?

Both carry Cursor's pstack skills to other agents. pstack-claude pre-translates the tool names for Claude Code and adds a routing hook, while the backnotprop mirror is a plain skills folder you install with the skills CLI.

Sources: GitHub repository (stars, license, file tree, last push Oct 28, 2026). Review based on the README and SKILL.md files as published.