Add a path without breaking the inherited one.
The original extraction workflow remains available. The new corpus workflow is separate, with documentation explaining which path fits which task.
04 / DOCUMENTATION ENGINEERING / OPEN-SOURCE DERIVATIVE
Corpus to Skill
Combine sources. Keep their differences.
A collection of documents can be useful guidance for an AI tool. It can also contain disagreements. I extended an existing document workflow so source-linked claims and their relationships survive the move into reusable guidance.
MY CONTRIBUTION
Extension design, corpus workflow, compatibility, lineage documentation, and evaluation. Built with AI coding assistance; derived from Book to Skill by Virgilio Junior.
THE RESULT
A bounded local text and Markdown corpus workflow alongside the inherited document extraction path.
ORIGINAL CC0 FIXTURES / QUEUE DESIGN
Read the condition behind the advice.
A curated illustration using the project’s fictional source fixtures. This page does not execute the corpus engine or present a live generated result.
SOURCE A / BOUNDED QUEUE OPERATIONS
Prioritize service stability.
INPUT / OBSERVATION
“Operators should drop requests when the queue is full if service stability is the objective.”
RECORDED RESULT
Condition: a full queue Objective: service stability Advice: drop requests
The source connects its recommendation to an explicit objective. That condition belongs beside the advice.
THE PROBLEM
What needed to change.
A polished summary can hide where an assertion came from or erase the conditions that make it true. I wanted the combined output to retain source references and disagreement instead of quietly presenting one synthesized answer as settled fact.
THE DECISIONS
Keep claims connected to passages.
Source references travel with the derived claims. A reader can move from the combined output back to the material behind it.
Make disagreement useful.
Relationship labels preserve possible agreement and tension between sources. Conditions stay visible, so conflicting queue advice can be understood in terms of stability, spare capacity, and request completion.
CURRENT STATE
What works today.
The public derivative includes architecture and maturity documentation, a deterministic offline corpus build, and original fictional queue fixtures for inspecting source relationships.
Relationship labels are heuristic. The tool is not a fact checker. Rich-document corpus adapters and live or model-based corpus integrations are not complete.