Analyzes skill content to determine model tier requirements.

Uses static analysis of the skill body and metadata to determine what model capabilities are needed. This runs at install time and when agents create new skills, injecting model_tier into the frontmatter if not already set.

Model Tiers

• :primary — Needs a capable model: tool use, complex reasoning, multi-step workflows
• :lightweight — Can run on a fast/cheap model: simple lookups, mechanical processes
• :any — No strong preference, works with either

Analysis Signals

Signals that push toward :primary:

• Tool calls referenced in the body (e.g., tool names, use_tool, function calls)
• Multi-step workflows (numbered steps, decision trees)
• Complex reasoning language (analyze, evaluate, decompose, synthesize)
• Delegation/coordination patterns (send_message, marketplace, sub-agent)
• SOP type with many steps

Signals that push toward :lightweight:

• Pure reference/documentation content
• Simple lookup or formatting tasks
• No tool usage mentioned
• Short body with no conditional logic