pi-backup

Multi-step deep research workflow for complex topics. Use when user asks for 深度调研、全面报告、thorough analysis、multi-source synthesis, or needs citations + structured conclusions.

Convert Markdown documents into beautiful, readable single-file HTML web pages and deploy them to the standalone web report site. Use when the user asks to "turn this markdown into a web page", "make this report into a web page", "deploy this as a web page", "render this markdown nicely", "把这个 md 转成网页", "生成网页报告", or any request to convert markdown/report content into a hosted, styled HTML page for easy reading and sharing.

Bridge between Discord and Pi. Allows managing Discord interactions, heartbeat tasks, and research reports. Use for managing the Discord bot, sending notifications to Discord channels, or checking bot status.

Guidelines for delegating all software engineering work to Codex CLI. Use non-interactive execution and ingest only final results to keep the main agent context clean.

Best practices for building agent-powered applications using pi-mono SDK (@mariozechner/pi-ai, @mariozechner/pi-agent-core, @mariozechner/pi-coding-agent). Use when the task involves: designing agent application architecture, embedding agent capabilities into web/backend services, writing system prompts for agent executors, implementing security for autonomous agents, building agent-as-backend systems, integrating pi-mono SDK into existing applications, creating Extensions or Skills for pi, or any development where an LLM agent loop drives real-world actions (deployment, infrastructure, data processing) rather than conversation.

Guide for extending pi coding agent following official standards. Use when creating extensions (custom tools with TypeScript), skills (reusable workflows with bundled resources), or prompt templates. Ensures compliance with pi extension API, skill specification (agentskills.io), directory structure, and progressive disclosure patterns.

Autonomous AI agent loop for iterative PRD-driven development. Generates PRDs, converts to prd.json, and executes user stories one at a time with quality gates, progress tracking, and pattern consolidation. Use when asked to: create a PRD, plan a feature, run ralph, autonomous development loop, iterative implementation, execute prd stories, or build a feature end-to-end autonomously.