bmad-distillator

Compress source documents into lossless token-efficient distillates for LLM consumption.

Updated May 7, 2026
One-click install
npx skills add https://github.com/jgabriellima/deep-agent-skill --skill bmad-distillator-jgabriellima
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: bmad-distillator
Source: https://github.com/jgabriellima/deep-agent-skill/tree/main/skills/bmad-distillator
Command: npx skills add https://github.com/jgabriellima/deep-agent-skill --skill bmad-distillator-jgabriellima

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Feeding large sets of source documents (briefs, discovery notes, research reports) into LLM workflows wastes tokens on prose, transitions, and repetition, while naive summarization loses critical facts, decisions, and constraints. This Skill produces lossless, hyper-compressed distillates that preserve every fact while stripping everything an LLM does not need. ## Core Features & Use Cases - Lossless Compression: Extracts every fact, decision, constraint, and relationship from source documents into dense thematic bullets, with completeness verification against source headings and named entities. - Semantic Splitting: Automatically splits large distillates into self-contained section files with an index when content exceeds token thresholds or a user-specified token budget. - Round-Trip Validation: Optionally reconstructs the original documents from the distillate alone and produces a gap/hallucination report to prove no information was lost. - Use Case: Before creating a PRD, distill a product brief plus its discovery notes into a single ~1,500-token distillate that the PRD workflow consumes as its sole context input. ## Quick Start Distill the documents in my product-brief folder into a single distillate optimized for PRD creation, and run the round-trip validation to confirm nothing was lost.

Frequently Asked Questions about bmad-distillator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I compress documents for LLM context without losing information?▼

Use lossless distillation instead of summarization: extract every fact, decision, and constraint into dense thematic bullets while stripping prose and redundancy. This Skill verifies completeness against source headings and named entities, with up to two targeted fix passes for gaps.

What is the difference between a distillate and a summary?▼

Summaries are lossy and drop details for readability. A distillate is lossless compression optimized for LLM consumption, preserving every fact, decision, rejected alternative, constraint, and open question while removing only human-oriented overhead like transitions and rhetoric.

How do I verify a compressed document lost no information?▼

Run the round-trip validation with the --validate flag. A separate reconstructor agent rebuilds the source documents from the distillate alone, then a semantic diff against the originals produces a report listing gaps and hallucinations.

Can I distill large document sets within a token budget?▼

Yes. Provide a token_budget and the Skill triggers semantic splitting: content is divided at natural topic boundaries into self-contained section files plus an index, rather than arbitrary size-based chunks. Sections target roughly 3,000-5,000 tokens each.

What file formats does document distillation support?▼

The analysis script accepts Markdown, text, YAML, and JSON files via file paths, folder paths, or glob patterns. It recursively scans folders while skipping directories like node_modules, .git, and virtual environments.

When should I not use round-trip validation?▼

Skip it for routine distillation because it adds significant token cost. Reserve it for critical documents where information loss is unacceptable, and note it requires subagent support since the main agent has already seen the originals.