Multi-LLM Orchestrator

Cross-model orchestration skill for Claude Code. Use multiple LLMs in a single workflow with dynamic routing and human gates between steps.

How It Works

Claude Code acts as the conductor — managing flow, calling external LLMs via scripts/orchestrate.py, and presenting human gates between steps.

Default flow:

GPT-5.4 (Analysis) → Claude Opus (Counter-analysis) → You (Gate) → Claude Code (Execute)

Any step can be rerouted at any gate. You're always in control.

Models

Alias	Model	Best for
gpt5	GPT-5.4	Analysis, structured critique, scoring
opus	Claude Opus 4.6	Counter-analysis, deep reasoning, implementation
sonnet	Claude Sonnet 4	Faster/cheaper Claude tasks
gemini	Gemini 2.0 Pro	Validation, cross-referencing, second opinions
gpt4o	GPT-4o	Legacy fallback

See references/model-profiles.md for detailed strengths, weaknesses, and routing heuristics.

Setup

1. Install the skill

Copy the skill directory into your Claude Code skills location, or clone this repo.

2. Set API keys

Set in your shell profile (~/.zshrc, ~/.bashrc, etc.):

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_AI_API_KEY="AIza..."

3. Verify

python scripts/orchestrate.py check

See references/config-setup.md for advanced configuration (config file, custom models, new providers).

Gate Protocol

At every step, you get:

Action	Effect
Approve	Proceed to next step with default model
Bounce	Return output to the originating model with feedback
Redirect	Send output to a different model
Adjust	Modify the prompt and re-run
Abort	Stop the workflow

File Structure

SKILL.md                          # Skill definition
scripts/orchestrate.py            # Python script for external LLM calls
references/
  workflow-templates.md           # Pre-built workflows and task file schema
  model-profiles.md               # Model strengths/weaknesses
  config-setup.md                 # API key and config setup
  api-patterns.md                 # Script internals (for debugging)

License

MIT