template-enrichment-beats-llm-for-bulk-skill-substance
Generating substantive invariant content for 69 skills (504 ceremonial sections) via a template-based Bash/Python script reduced warnings from 504 to 0 in a single run — far faster and cheaper than per-skill LLM generation. The script contextualizes content by skill layer (1–7, Meta, Overlay) and description, replacing only sections with <15 tokens. This pattern generalises: whenever >20 files need structurally similar content, a deterministic template script outperforms an LLM agent loop.