Weave production-grade LLM prompts with cache boundaries, tool injection, and token budgeting. Reverse-engineered from Claude Code's 7-layer prompt architecture.
A toolkit for batch LLM API calls driven by YAML configuration.