Pi Agent Kit

A collection of skills and extensions for AI coding agents, compatible with pi-coding-agent and related agent runtimes.

Installation

# Install from git
pi install git:github.com/aldoborrero/pi-agent-kit

# Or from local clone
pi install ./pi-agent-kit

# Test a single extension
pi -e ./extensions/notify/notify.ts

Repository Structure

pi-agent-kit/
├── agents/           # Agent definitions (6 agents)
├── extensions/       # TypeScript extensions (33 local + 3 npm)
├── prompts/          # Workflow templates (4 prompts)
├── skills/           # Markdown skills (portable across agent runtimes)
│   ├── ast-grep/     # AST-based structural code search
│   ├── kagi-search/  # Privacy-focused search
│   ├── pexpect-cli/  # Interactive CLI automation
│   ├── searxng-search/ # Self-hosted SearXNG search skill
│   └── superpowers/  # 13 advanced workflow skills
├── themes/           # Color themes (lavender)
└── packages/         # Nix packages (pexpect-cli, searxng-search)

Pi Extensions

TypeScript extensions for pi-coding-agent. 33 local extensions + 3 external npm packages.

Code Intelligence & Review

Extension	Description
ast-grep	Structural code search using AST patterns with pattern, rule, and inspect modes
tuicr	Interactive code review tool — agent launches tuicr, captures your review feedback
diff	/diff command — interactive diff viewer (tuicr/delta/git)
codex	OpenAI Codex integration — /codex review, /codex rescue, adversarial reviews
pi-lsp-extension	LSP integration — diagnostics, hover, go-to-definition, references, symbols, rename

Safety & Guardrails

Extension	Description
sandbox	OS-level sandboxing for bash — /sandbox on/off
permission-gate	Confirms before dangerous bash commands (rm -rf, git push --force, sudo, etc.)
git-checkpoint	Git stash checkpoints at each turn so /fork can restore code state
bitwarden	Secure vault access via rbw — bw_get/bw_list tools with password masking

Session & Context Management

Extension	Description
context	/context TUI dashboard — token usage bar, loaded extensions/skills, session cost
handoff	/handoff <goal> — transfer context to a new focused session
btw	/btw <question> — ephemeral side question (no tools, no context pollution)
recorder	Record all session activity to SQLite for performance tracking and analytics

Workflow & Automation

Extension	Description
plan-mode	/plan — read-only exploration with subagent orchestration and plan file output
loop	/loop 5m <prompt> — periodic polling and monitoring on a schedule
until	/until tests — repeat until condition met (TDD, iterative fixes)
subagent	Delegate tasks to specialized subagents with isolated context (single, parallel, chain)
pi-interactive-shell	Full PTY emulation for interactive CLIs — user can observe and take over anytime

Search & Web

Extension	Description
web-tools	High-level web_search and web_fetch tools — preferred interface for web search, webpage fetching, and source gathering (Exa, Brave, SearXNG)
github-search	Search code, issues, and PRs on GitHub via gh CLI
exa-search	Backend-specific Exa search integration used by higher-level web tooling
brave-search	Backend-specific Brave search integration used by higher-level web tooling
jina	Backend-specific webpage-to-markdown fetcher used by higher-level web tooling

Model Providers

Extension	Description
openrouter-provider	Kimi K2.5, Gemini, Grok, Devstral, ByteDance — one API key for all
groq-provider	GPT-OSS 120B/20B, Qwen3, Llama 4 Scout — ultra-fast LPU inference
together-provider	Llama, DeepSeek, Qwen, Kimi, GLM, Mistral — 25+ open-source models

Environment & Integration

Extension	Description
direnv	Auto-load direnv environment variables — /direnv to reload on demand
walkie	Telegram bridge for mobile — bidirectional messaging, voice, photos, draft streaming
pi-mcp-adapter	Token-efficient MCP (Model Context Protocol) adapter

Input & Voice

Extension	Description
voice	Toggle-to-record speech-to-text — Groq, OpenAI, or local Whisper daemon. Ctrl+Alt+V
inline-bash	Expand !{command} patterns in prompts
questionnaire	Structured multi-question UI with options and free-text input

UI & Commands

Extension	Description
notify	Desktop notifications when the agent finishes
footer	Custom footer with git branch, context usage, and extension statuses
skill-namespaces	/superpowers:* namespaced skill commands in autocomplete
exit	/exit command — alias for /quit
git-commit-context	/commit command with git status/log context injection

Skills

Markdown-based instructions that teach AI agents how to use external tools. Compatible with pi-coding-agent and other runtimes via the Agent Skills standard.

Tool Skills

Skill	Description
/ast-grep	Structural code search using AST patterns — find code by structure, not text
/pexpect-cli	Automate interactive CLI programs (SSH, databases, editors) with pexpect and pueue
/kagi-search	Privacy-focused web search via Kagi with Quick Answer support
/searxng-search	Query a self-hosted or public SearXNG instance with normalized, LLM-friendly output

Superpowers Skills

Skill	Description
/superpowers-brainstorming	Explore intent, requirements, and design before implementation
/superpowers-writing-plans	Create detailed implementation plans with bite-sized tasks
/superpowers-executing-plans	Execute plans task-by-task with review checkpoints
/superpowers-subagent-driven-development	Execute plans by delegating independent tasks to subagents
/superpowers-dispatching-parallel-agents	Run 2+ independent tasks in parallel without shared state
/superpowers-test-driven-development	Write tests before implementation code
/superpowers-systematic-debugging	Structured debugging with root cause analysis
/superpowers-verification-before-completion	Run verification commands before claiming work is complete
/superpowers-requesting-code-review	Request formal code review before merging
/superpowers-receiving-code-review	Process code review feedback with technical rigor
/superpowers-using-git-worktrees	Create isolated git worktrees for feature work
/superpowers-finishing-a-development-branch	Guide completion of development work (merge, PR, or cleanup)
/superpowers-writing-skills	Create, edit, or verify skills before deployment

Agents

Agent definitions for pi-coding-agent. Each agent is a lean system prompt that loads skills at runtime.

Agent	Description
brainstormer	Collaborative design dialogue — explores ideas before implementation
debugger	Systematic debugging specialist with root cause analysis
planner	Creates bite-sized implementation plans from context and requirements
reviewer	Code review for quality, security, and maintainability
scout	Fast codebase reconnaissance that returns compressed context for handoff
worker	General-purpose implementation with TDD, verification, and debugging

Prompts

Workflow templates that orchestrate agents into chains via the subagent extension.

Prompt	Description
/brainstorm	Collaborative design dialogue
/debug	Systematic debugging — scout gathers context, debugger investigates
/full-cycle	Full lifecycle — scout → planner → worker → reviewer → worker
/review	Standalone code review of recent changes or specified files

Packages

Package	Description
pexpect-cli	Persistent pexpect sessions via pueue — server/client CLI for interactive automation
searxng-search	CLI for querying SearXNG instances with normalized, LLM-friendly output

Development

nix develop               # Enter dev shell
nix fmt                   # Format code
nix build .#pexpect-cli   # Build pexpect-cli
nix build .#searxng-search # Build searxng-search

License

MIT