hybrid-agent-detector

Hybrid-agent-detector analyzes agent prompts and files under .claude/agents to identify execution patterns like file operations, data processing, and API calls that should be delegated to skills. It returns a structured JSON report with a hybrid_score and specific execution patterns found, helping you spot where orchestration logic bleeds into direct work.

The detector scans for direct execution across Read, Write, Edit, Bash, and API call tools within agent prompts. It distinguishes orchestration patterns from execution work, flagging cases where agents handle file operations, data transformation, or external calls instead of delegating to specialized skills.

Run the detector to get execution_patterns and recommendations in the JSON output. The results guide you to consolidate flagged execution logic into dedicated skills, leaving the agent to orchestrate skill calls. Follow the recommendation field to identify which work should migrate into new or existing skills.

Yes. The detector examines orchestration vs execution patterns across diverse project contexts by analyzing agent definitions in .claude/agents. It works regardless of your project structure, scaling from single agents to multiple orchestration patterns.

The detector returns a structured JSON with agent entries, a hybrid_score rating severity, execution_patterns listing found work, orchestration_present status, and a recommendation for refactoring. This structured format lets you prioritize which agents need consolidation into skills.