bmad-distillator

Lossless document distillation preserves every fact, decision, constraint, and relationship from multiple source documents while removing overhead, ensuring high-fidelity inputs for downstream LLM workflows like PRD creation and architecture synthesis.

The distillation process uses semantic splitting and stage-based processing across Analyze, Compress, Verify & Output, and optional Round-Trip Validate stages to handle large source sets and deliver predictable token footprints for downstream tasks.

Yes, the Skill provides token estimation and strict frontmatter metadata alongside the compressed distillate, allowing you to predict and control the token footprint before executing downstream LLM tasks.

Yes, lossless distillation is suitable for PRD creation and architecture synthesis, compressing multiple source documents into a token-efficient format that maintains faithful fidelity for long-context LLM workflows.

The optional Round-Trip Validate stage verifies that document compression did not lose any facts or constraints by validating the generated distillate against the original source documents.

No, specifying a downstream consumer is optional when generating a lossless distillate from source documents, but providing it helps tailor the compression output to meet specific downstream LLM workflow requirements.