cullenmcdermott/llm-orchestrator

LLM Orchestrator

Overview

This skill provides shared infrastructure for multi-LLM orchestration workflows. It enables discovering which LLM CLI tools are installed, assessing the complexity of code changes to determine reviewer allocation, and generating structured prompts for different review/analysis tasks.

Available Scripts

scripts/discover_llm_clis.py

Detect installed and authenticated LLM CLI tools. Returns JSON with availability status for each supported CLI.

Usage: uv run scripts/discover_llm_clis.py

Output format:

{
  "available": ["claude"],
  "unavailable": ["cursor-agent", "llm", "gemini", "aider"],
  "details": {
    "claude": {"installed": true, "path": "/usr/bin/claude", "version": "2.1.36"}
  }
}

scripts/assess_complexity.py

Analyze a git diff to determine change complexity and recommend reviewer allocation.

Usage: uv run scripts/assess_complexity.py [--diff-args "HEAD~1"]

Defaults to staged changes if no diff args provided.

Complexity levels: | Level | Criteria | Recommended Reviewers | |-------|----------|-----------------------| | small | <50 lines, 1-2 files | architect + stylist (2) | | medium | 50-200 lines, 3-5 files | + tester (3) | | large | 200-500 lines, 5+ files | + perf + external (5) | | critical | 500+ lines OR touches auth/crypto/infra | all 6 + external (7) |

Output format:

{
  "complexity": "medium",
  "lines_changed": 120,
  "files_changed": 4,
  "touches_sensitive": false,
  "recommended_reviewers": ["reviewer-architect", "reviewer-stylist", "reviewer-tester"],
  "summary": "Medium change: 120 lines across 4 files"
}

scripts/enhance_prompt.py

Generate structured prompts for different analysis tasks. Takes a task type and optional context, returns a formatted prompt.

Usage: uv run scripts/enhance_prompt.py <task_type> [--context "additional context"]

Supported task types: review, security, test-gen, explain, commit-msg, adr

Integration Pattern

Sub-agents and commands should use these scripts as building blocks:

1. Discovery (discover_llm_clis.py) — Called at the start of multi-model workflows to determine which CLIs are available for dispatch.
2. Complexity (assess_complexity.py) — Called by /multi-review to scale the number of reviewer agents.
3. Prompts (enhance_prompt.py) — Called by sub-agents and commands to get consistent, high-quality prompts for each task type.