skills

Personal agent skills for the open skills.sh ecosystem. Includes pstack, a harness-neutral fork of Cursor's pstack.

Install

# all skills in this repo
npx skills add MondayNotAGoodDay/skills

# one skill
npx skills add MondayNotAGoodDay/skills@eli5
npx skills add MondayNotAGoodDay/skills@poteto-mode

List without installing:

npx skills add MondayNotAGoodDay/skills --list

Browse on skills.sh: https://skills.sh/MondayNotAGoodDay/skills

pstack

pstack is a set of rigorous engineering workflows: poteto-mode routes a task to a playbook, and the other skills (how, why, arena, swarm, interrogate, the principle-* skills) run as the steps need them. Read the guide.

Upstream is written for Cursor. This fork runs on any harness (opencode, Devin, Cursor, Claude Code, and others) with one setup step.

Setup

Run once per harness:

/setup-pstack

Skills never name a harness tool or a model. They say "spawn role X with read-only access" and "ask the user". /setup-pstack finds out how your harness does those things by reading its own tool schemas, picks a model per role and a reasoning budget, and writes a short pstack profile into the harness's standing instructions. Where a harness binds the model to an agent definition instead of the spawn call (Devin does), it generates the agent files. It then spawns real subagents with the finished recipe to prove it works.

Without a profile, skills still run. They spawn a general-purpose subagent with no model override, or run roles one after another if the harness cannot spawn.

Entry points

Skill	When to use
setup-pstack	Adapt pstack to this harness. Run first.
poteto-mode	Default entry for any task that needs rigor.
poteto-help	New to pstack, or unsure which skill fits.
how / why	How a subsystem works / why it was built that way.
architect / arena / swarm	Settle a shape / N parallel attempts / N parallel workers.
interrogate	Multi-reviewer review of a branch or PR.
ketchup	Catch up on what changed in this chat since your last message, in plain words and small visuals, with every waiting call listed so you can decide from the reply alone.
ketchup-video	The /ketchup catch-up as a short silent video of slides, with the calls waiting on you as text under it.
in-your-own-words	Play back the goal and the problem before the work starts.
sharpen	Rewrite a prompt deeper, not bigger, before you send it. Hands back the rewrite, leaves your choices open, and doesn't do the task.
create-formal-verification-skill	Prove one property tests rarely reach, and leave a project skill that rechecks it.
maintain-formal-verification-skill	After covered code changes, bring the formal verification back in line and say whether a failure is drift or a bug.
statechart	Draw what the user sees as a Mermaid statechart from a TLA+ model, with each bug as a numbered path. Also turn an agreed chart into a TLA+ skeleton, or audit the code against a spec.

The rest (blast-radius, recall, reflect, tdd, unslop, no-comments, the 24 principle-* skills, and others) are in skills/.

What differs from upstream

• Spawn blocks use role / access / run instead of Cursor's Task parameters. Model slugs are gone from every skill. Roles and their model classes live in setup-pstack/references/roles.md.
• poteto-agent and Comment Sicko are prompt files (poteto-agent, comment-sicko), not Cursor agent definitions.
• Transcript, loop, and question tools are capabilities in the profile, not named tools.
• Not ported: make-bot-ui and the Benny automation pack. Both depend on Cursor Automations.

UPSTREAM.md records the upstream commit this fork follows. Prompt prose tracks that commit. Spawn blocks stay on the profile.

Other skills

Skill	When to use
eli5	Dead-simple picture explainer (/eli5 <topic>)
show-me	Visual diagrams / code-shape sketches for the current topic
html-plan	An implementation plan as one reviewable HTML page (/html-plan <what to build>)

Layout

skills/
  <skill-name>/
    SKILL.md
    references/   optional: long reference the skill reads on demand
    runtime/      optional: scripts the skill runs
    examples/     optional: a worked example
docs/guide/        pstack guide
third_party/       upstream license

Each SKILL.md has YAML frontmatter with name (equal to the directory name) and description. Most skills also set disable-model-invocation: true, so they run only when named.

License

MIT. pstack is derived from cursor/plugins, Copyright (c) 2026 Lauren Tan, MIT, see third_party/pstack/LICENSE. html-plan is from anthropics/claude-plugins-community, Copyright (c) 2026 Thariq Shihipar, MIT, see third_party/html-plan/LICENSE. eli5 adapted from the same community repo; show-me from humanlayer/skills. UPSTREAM.md records every upstream commit.