A process-discipline skills bundle for pi — the minimal terminal coding agent by Mario Zechner.
Inspired by Claude Code's superpowers plugin, this package ships a set of Agent Skills that nudge pi toward better engineering habits: brainstorming before building, structured debugging, TDD, written plans, verification before claiming "done", and isolated work via git worktrees.
Skills auto-load and the model picks the right one based on your request — no slash-commands required.
# From npm (once published)
pi install npm:pi-superpowers
# From git
pi install git:github.com/krone9/pi-superpowers
# From a local clone (for development)
pi install /absolute/path/to/pi-superpowersAdd -l to install into a project's .pi/settings.json instead of global.
Verify installation:
pi listYou should see the package listed; on next pi startup the skills appear in the header.
Each skill is a markdown file with name + description frontmatter. Pi loads the descriptions into its system prompt at startup; full instructions are pulled in on-demand when the model decides a skill applies (or when you force one with /skill:name).
| Skill | When pi will reach for it |
|---|---|
using-superpowers |
Bootstrap — instructs the model to check skills before responding |
brainstorming |
New feature, vague goal, "what should we build…" |
writing-plans |
Multi-step changes that should be planned before coding |
test-driven-development |
Adding behavior that can be tested |
systematic-debugging |
Bug reports, unexpected behavior, regressions |
verification-before-completion |
Before claiming a task is done |
using-git-worktrees |
Isolated feature/experiment branches |
dispatching-parallel-agents |
2+ independent tasks that can run concurrently |
writing-skills |
Authoring a new skill |
These mirror Andy's ~/.claude/skills/ set — process/logging/git helpers. Pi auto-loads them by description; you can also force-load any with /skill:<name>.
audit-structure, brief, changes, commit, debt, dependencies, log, nextsteps, optimise, restart, review, rollback, rules, scaffold, security, staging, status, wrapup
Skills are plain markdown — fork, edit, or add your own under skills/<your-skill>/SKILL.md. The name field must match the parent directory.
Pi runs many providers. Forced skill-checking is prompt persuasion, not a tool the model is trained on. Capable models (Claude, GPT-4-class, Gemini Pro) follow the bootstrap reliably; smaller models may ignore it. The skills themselves work everywhere — only the always-check-first nudge depends on model strength.
Concept and skill names adapted from the Claude Code superpowers plugin by Anthropic. Reimplemented for pi's skills system. Not affiliated with Anthropic or pi.dev.
MIT