pi-mini-orchestrator

Orchestrates claude and mini sub-agents with scoped MCP servers, provider fallback chains, and git worktree isolation.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/nbiish/tts-mcp --skill pi-mini-orchestrator-nbiish
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: pi-mini-orchestrator
Source: https://github.com/nbiish/tts-mcp/tree/main/.agents/skills/pi-mini-orchestrator
Command: npx skills add https://github.com/nbiish/tts-mcp --skill pi-mini-orchestrator-nbiish

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running multiple AI coding agents in parallel often leads to API failures, uncontrolled tool access, and conflicting file changes. This Skill coordinates sub-agent dispatches so each task gets only the MCP servers it needs, automatically recovers from provider errors, and isolates work in separate git worktrees. ## Core Features & Use Cases - Fixed Agent Rotation: Dispatches tasks in a strict claude → mini rotation, using claude for reasoning-heavy work (audits, reviews, design) and mini for execution (fixes, tests, migrations). - Provider Fallback Chain: Automatically retries failed API calls across an ordered chain of eight providers (modal → nvidia → nebius → opencode → zai → wafer-serverless → openrouter → zenmux), then falls back to the other agent. - Scoped MCP Configuration: Dynamically writes ~/.claude.json or a temporary mini YAML config so each sub-agent receives only the MCP servers relevant to its task, with cleanup via trap handlers. - Worktree Isolation: Assigns one git worktree per dispatch, then merges, verifies, and cleans up results centrally. - Use Case: Dispatch a security audit to claude with github, tavily-search, and gitnexus MCP servers in one worktree, while mini writes a database migration with the postgres MCP server in another, then merge both after review. ## Quick Start Ask the orchestrator to dispatch a code review task to the next agent in rotation with the appropriate MCP servers and an isolated git worktree.

Frequently Asked Questions about pi-mini-orchestrator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multiple AI coding agents in parallel without conflicts?▼

Assign each sub-agent its own git worktree so concurrent agents never write to the same tree. The orchestrator reviews each worktree's changes, merges them back, verifies the result, and removes the worktree after completion.

How do I handle API rate limits and provider failures with Claude Code?▼

Use an ordered provider fallback chain that retries the next provider on rate limits, auth failures, or timeouts. This Skill cycles through modal, nvidia, nebius, opencode, zai, wafer-serverless, openrouter, and zenmux with a 2-second cooldown between attempts.

How do I configure MCP servers for Claude Code dynamically?▼

Write a temporary ~/.claude.json containing only the required mcpServers entries before launching claude, backing up any existing config first. A trap handler restores the original file or deletes the temporary one when the dispatch exits.

Which MCP servers should I use for a security audit or PR review?▼

For security audits, use github, tavily-search, and gitnexus with claude as the agent. For PR reviews, github and gitnexus are recommended, also dispatched to claude since it handles reasoning-heavy tasks.

When should I use mini instead of claude for a coding task?▼

Use mini for execution-focused work such as bug fixes, unit tests, database migrations, and boilerplate generation. Claude is preferred for research, audits, design, refactoring, and review tasks that need deeper reasoning.

What are the limitations of this multi-agent orchestration approach?▼

Dispatch aborts after both agents exhaust all eight providers, totaling 16 attempts, with a 5-minute timeout per attempt. It also requires API keys supplied through environment variables and depends on the claude and mini binaries being installed.