Personal monorepo of pi extensions, themes, and skills.
extensions/ # Auto-discovered .ts extension files (subdirs with index.ts)
themes/ # Auto-discovered .json theme files
skills/ # Auto-discovered SKILL.md folders
pi install git:github.com/dfontana/pi-extensionsEach extension has its own README covering configuration, provided tools/commands, and limitations.
| Extension | Purpose |
|---|---|
| claude-marketplace | Pull skills from Claude Code plugin marketplaces into Pi, auto-synced on a schedule |
| copy-response | /cp to copy the last response or a code block, whitespace-normalized so formatting survives pasting |
| helix-mode | Helix-style modal editing for the input box (on by default; /helix to toggle) |
| improved-footer | Footer with jj bookmark support and accurate OpenRouter cost tracking |
| mcp | Lean MCP client: .mcp.json servers behind a single mcp proxy tool, /mcp panel |
| no-git-for-jj | Blocks git commands in jj repos, suggesting the jj equivalent |
| notify | Bell, desktop notification, and Zellij tab dot when Pi is ready for input |
| pane-control | Multiplexing tools for opening, driving, reading, and closing Kitty or Zellij panes |
| prompt-stash | Stash/restore the prompt editor with Alt+Shift+S |
| model-query | model_query tool: resolve one usable model by deterministic policy |
| review-model-selector | select_review_model tool: deterministic adversarial-reviewer selection |
| scoped-tools | YAML-specified bash commands as validated agent tools with hidden, call-time computed parameters |
| subagent | Lean synchronous one-call delegation to global agents, with sibling-call parallelism |
| web-access | web_search and web_fetch tools backed by configurable model providers |
- quick-review — lightweight review/fix workflow for small, routine, low-risk, or clearly scoped changes: one reviewer covers completeness, correctness, duplication, and simplicity in each bounded round. This is the default for ordinary review requests.
- run-review — full adversarial review/fix workflow for explicitly deep, comprehensive, multi-agent, broad, or high-risk reviews: four sibling subagent investigations emitted in one response, an independent synthesis reviewer, and a bounded fix/re-review loop.
- run-plan — consolidates review feedback, delegates implementation to a persistent large-context agent, then invokes the full
run-reviewworkflow with that implementer as the fixer.
A repo-local web-access-smoke-test skill (in .pi/skills/, not shipped with the package) provides an end-to-end check of the web-access tools after code changes.
quick-review, run-review, and run-plan require this package's subagent and review-model selection tools; they preflight those tools and return enablement guidance rather than silently degrading.
/skill:quick-review
/skill:quick-review file=extensions/example.ts
/skill:run-review scope=branch base=main thinking=xhigh
/skill:run-plan <review feedback and optional instructions>
- rose-pine-dawn — Rosé Pine Dawn, a warm light theme with soft pastels.
- npm:@juicesharp/rpiv-ask-user-question
- npm:@tintinweb/pi-tasks
- npm:@plannotator/pi-extension
- git:github.com/dfontana/pi-extensions