Purpose
SourceReady NAS is an Uliya orchestration Skill that turns a creator's scattered notes, transcripts and media metadata on one topic into a short prep brief where every statement points back to a real file.
User and problem
Solo creators keep notes, gear lists, shot lists and old transcripts across NAS folders. Before filming, they reread everything by hand and still miss outdated or conflicting details.
Solution
The user names a topic, up to three folders, a date meaning and an optional output folder. The Skill returns a summary whose sentences cite claim IDs, a material ledger, a claim ledger with source locations, and a list of gaps and conflicts.
Core flow
Confirm inputs, discover candidates without reading content, check that every path stays inside the named folders, confirm a read set of at most fifteen content files (media metadata separate), read with citations, assemble the brief, and save only after the user chooses and the platform approves.
What makes it different
It separates file time from content date, flags conflicts instead of averaging them, treats instructions inside files as data, and by design asks before any write, saving revisions as new files rather than replacing earlier output.
Mapping to NAS steps
Discovery, direct text reading and media metadata go to file-agent, cited document reading to document-agent, and saving to knowledge-agent. The parent Skill performs no NAS operation itself; it checks scope and assembles the brief from returned facts.
Technical approach, inputs, permissions, model and data
One SKILL.md plus three Markdown references, no code. It uses ug-filesystem, ug-filemgr, ug-document and ug-knowledge. It does not circumvent platform file permissions, though their enforcement is unverified; saving also needs workflow approval. Inference is whatever Uliya is configured with; its deployment and retention are unknown to me, so no on-device or privacy claim is made.
Current status
A locally authored package with synthetic demo files and a scenario walkthrough. The official local static checker passed. Tests on a real NAS are pending.
Tests and edge cases
Device test 1: brief from a dedicated folder with a known conflict. Device test 2: save to an output folder that already holds an earlier brief. Edge cases include partial reads, empty results, embedded instructions, scope leaks and approval denial.
Limitations
No video frame analysis, new OCR, new transcription or web search in this version. Limits and the no-overwrite policy are instructions, not code-enforced or device-verified guarantees, and listings cut at a limit are reported as incomplete.
New work
Created new for this hackathon in November 2026.
AI development tools
Claude Code (Anthropic) drafted the Skill text, references, synthetic demo content, documents and packaging helpers. Codex (OpenAI) helped with planning, independent review and draft helper code. Submission and demo material is public documentation, the official kit, or synthetic content; no patient or customer data was used. Tool use is subject to each provider's terms.
Third-party material and licenses
The official Uliya Skill development kit 0.7.7 (catalog 2.2.0, from the hackathon starter kit page) was used only as a specification and local checker and is not redistributed. No kit license is stated here; its use is subject to the hackathon rules. No third-party runtime code is included. No open-source license is claimed.
Data and scope
Demo files are synthetic. The Skill adds no external service; runtime inference is the user's Uliya configuration, whose data handling is not claimed here. It is not intended for medical or patient data and gives no medical advice. No measured time savings are claimed.
Built With
- markdown
- uliya
