Inspiration
Reports often combine documents that disagree or change. A concise answer is more useful when readers can see which source version supports each claim.
What it does
Source-Linked Brief describes a bounded NAS workflow for up to ten explicitly selected documents. It returns a brief, a claim/source/version ledger, disagreements and evidence gaps, with source rechecks and an optional approved save.
How it was built
The new Skill and core workflow were authored on 9 November 2026 using the official Uliya Skill Kit 0.7.3 and catalog 2.0.0. It delegates to the registered file, document and knowledge experts. OpenAI Codex assisted drafting, mapping and checks; AI review was checked against evidence. Vendor code, model weights, training data and private documents are not bundled.
Validation and next steps
The official local checker reports static-pass with zero errors or warnings. Independent checks confirm package structure, source identity, the proposal and rejection of invalid package mutations. These are static results. Two core cases and one permission/hostile-content edge case are planned for the assigned NAS if selected as a finalist. Installation, extraction quality, model behavior and saving require device validation; no summarization accuracy or operational NAS demonstration is claimed.
Author
Francisco Angulo de Lafuente - ORCID 0009-0001-1634-7063. Original materials remain the author's property, subject to the official competition evaluation and publicity permissions.
Research background
This project draws on my earlier work on AI memory and document retrieval. Unified-Holographic-Neural-Network, a research line begun in 2024, includes a persistent document index, exact source coordinates, learned query-to-passage associations and checked reversible memory export. ASIC-RAG-CHIMERA explores content-addressed storage for RAG, and CAJAL RAGFlow connector connects retrieved chunks to local-model document generation. These are background research references; their code and old submission materials are not bundled or reused as this new UGOS Skill. Their published experiments are not presented as performance results for this package.
