Community Extensions¶

Pi's community has built extensions that cover the major features found in competing agents. Here are the most relevant packages.

shitty-extensions (by hjanuschka)¶

The most comprehensive community package. Actively maintained.

pi install npm:shitty-extensions

Extension	Description	Equivalent in
`oracle.ts`	Get second opinions from other AI models	Amp Oracle
`handoff.ts`	Transfer context to new sessions	Amp Handoff
`plan-mode.ts`	Read-only exploration mode	Claude Code plan mode
`memory-mode.ts`	Save instructions to AGENTS.md	Persistent learning
`cost-tracker.ts`	Session spending analysis	Built-in in Amp/Claude
`clipboard.ts`	Copy text to system clipboard via OSC52	--
`ultrathink.ts`	Rainbow animated "ultrathink" effect	Fun
`loop.ts`	Conditional loops (by mitsuhiko/Armin Ronacher)	--
`flicker-corp.ts`	Authentic fullscreen flicker experience	Parody of Claude Code

Oracle Usage¶

Once installed, tell the agent:

"Ask the oracle to review this implementation"
"Use the oracle to debug this race condition"
"Have the oracle brainstorm alternative approaches"

The oracle sends the current context to a second model (configurable) and returns its analysis. It's read-only -- it advises but doesn't make changes.

Handoff Usage¶

/handoff now implement the authentication flow
/handoff execute phase one of the plan

Generates a focused prompt for a new session based on the current context and your goal.

pi-model-switch (by nicobailon)¶

Lets the agent switch models autonomously.

pi install npm:pi-model-switch

Configure aliases:

{
  "cheap": "google/gemini-2.5-flash",
  "coding": "anthropic/claude-opus-4-5",
  "budget": ["openai/gpt-5-mini", "google/gemini-2.5-flash"]
}

Usage:

"Switch to a cheaper model"
"Use Opus for this refactor"
"List available models"

pi-subagent-enhanced (by nicobailon)¶

Full sub-agent support with multiple execution modes.

Mode	Description
Single	`{ agent: "worker", task: "refactor auth" }`
Chain	Sequential tasks with `{previous}` placeholder
Parallel	Multiple tasks running simultaneously
Async	Background execution with notifications

Features: - Output truncation (configurable byte/line limits) - Debug artifacts (input, output, JSONL, metadata per task) - Session-scoped notifications

@marckrenn/pi-sub-core¶

Shared usage tracking across providers.

pi install npm:@marckrenn/pi-sub-core

Tracks usage for: Anthropic, OpenAI Codex, GitHub Copilot, Google Gemini, Antigravity, z.ai, AWS Kiro.

pi-skills (by Mario Zechner)¶

Mario's official collection of skills for common development tasks.

pi install git:github.com/badlogic/pi-skills

These are curated, first-party skills that follow pi's philosophy of progressive disclosure -- the agent loads them on-demand when relevant. Check the repo for the current list of available skills and their descriptions.

https://github.com/badlogic/pi-skills

Armin Ronacher's Extensions (referenced in his blog)¶

Armin has built several extensions he describes in his Pi writeup:

/answer -- Extracts questions from the agent's response, presents them in a structured UI, sends answers back
Custom to-do tracker -- Agent-specific local issue tracker with a tool interface
Various TUI widgets -- Dashboards, debugging interfaces

His philosophy: point your agent to an existing extension and say "build it like that, but with these changes."

Notable Community Integrations¶

Emacs frontend (dnouri/pi-coding-agent) -- Full Emacs mode with markdown rendering, streaming, branch navigation
OpenClaw -- Slack/Telegram bot built on pi's SDK
pi-mom -- Mario's autonomous Slack bot

Finding More Packages¶

# Browse on npm
# Search for keyword: pi-package

# Or check the Discord community server

The package registry at shittycodingagent.ai/packages lists community contributions (when npm registry is reachable).