Inspiration
As my grandmother's voice weakened, conversations with my family became shorter. That experience motivated my interest in communication interfaces. Care Log Digest explores a separate part of that problem: carrying a person's recorded request into the next shift without confusing an unrelated note with a completed response.
What it does
Care Log Digest is a Round 1 UGREEN Skill prototype described in Markdown. It delegates bounded file reading, evidence-based drafting, and reviewed saving to registered UGREEN experts. The draft keeps urgent request quotations, matching response records, and source citations visible. It does not provide medical advice, send alerts, or connect to a mouth-input application.
How we built it
I am entering individually. The workflow was developed during the hackathon using the official UGOS Skill Creation Tool, uliya-skill-kit 0.7.3-en and capability catalog 2.0.0. The parent delegates to file-agent, document-agent, and knowledge-agent; no executable scripts are included in the submitted ZIP. The workflow uses care request keys to distinguish acknowledgment from an explicit completion record, asks staff to review the draft, and requires a verified new-file save.
Development status
The Markdown workflow, log format, draft template, synthetic validation cases, and disclosures are prepared. The official local checker result is supplied in the preparation record. No UGREEN NAS runtime test or care-home deployment has occurred. Finalist testing must verify model routing, permissions, CSV coverage, content-only drafting, Japanese text, and saving without overwriting an existing file.
Challenges
A later note in the same room does not necessarily answer the resident's request. The design therefore requires an exact care request key and an explicit completion record, preserves the source evidence, and stops when that evidence is incomplete. Model instructions are not a guarantee of deterministic behavior, so review remains part of the workflow.
Accomplishments
The initial broad idea has been converted into an auditable workflow based on named, documented UGREEN capabilities. Its claims distinguish prepared materials from untested device behavior, and its examples contain only fictional records.
What is next
If selected, validate two core cases and an invalid-input edge case during the assigned NAS testing day. Record actual results and limitations, adapt only through supported capabilities, and retain the original Round 1 materials unchanged.
Built with
UGOS Skill Creation Tool; Uliya; Markdown; CSV; ug-filesystem; ug-document; ug-knowledge
AI disclosure
Anthropic Claude produced the supplied initial concept and preliminary draft. OpenAI Codex researched official documentation, rewrote the workflow and proposal, designed the matching and failure rules, prepared fictional cases, packaged the files, and ran static checks. No claim is made that the entrant has already reviewed the final text. Runtime inference depends on the UGREEN environment; its model route has not been validated, so unknown routing restricts testing to fictional data.
Third-party components / external data
The official UGREEN creation tool was used for development and is not redistributed in the Skill ZIP. No third-party runtime SDK, dataset, model weight, or API integration is bundled. Source and applicable terms are identified in the PDF and Skill disclosures. The Skill does not authorize external transfer of real private records.
Log in or sign up for Devpost to join the conversation.