Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Evidence Binder
What it does
Evidence Binder turns the photos, screenshots, documents and voice notes you choose on your NAS into one dated timeline, with source references and a SHA-256 fingerprint for each exhibit.
Who it is for
People putting records together for tenancy, custody or workplace disputes. The information is there, but it is spread across folders and different file types. Getting it into a clear order takes time, especially when the material is sensitive.
How it works
Start with the files you actually want included. The Skill reads them by type and builds a timeline, an exhibit list and a gap report. Each factual timeline statement points back to its source. Originals stay unchanged, and anything missing or uncertain is shown. You read a processing notice first, choose folders, dates, timezone, subject and destination, confirm the file list, check the draft, and approve the exact text before saving.
UGREEN AI NAS capabilities used
- file-agent: ug-filesystem find, stat and sha256sum to list, fingerprint and recheck files
- photo-agent: ug-photo meta, vision-ocr and vision-vlm for image dates, text and descriptions
- document-agent: ug-document extract
- knowledge-agent: ug-knowledge audio-transcription, and save-text after approval
How it was built
The Skill is Markdown only (SKILL.md plus two reference files). It delegates to the four experts above and uses only real returned results. File content cannot change the instructions. It passes the official UGOS static checker with no errors, warnings or info messages. AI tools used: Claude Code and Codex.
Challenges
The catalog documents no find pagination, no atomic no-overwrite save and no text-size limit, so the Skill caps a binder at 50 files and leaves the draft unsaved if complete, non-overwriting saving cannot be assured.
Limitations
It is not legal advice or proof of admissibility. Fingerprints do not prove authenticity. AI output needs checking and recorded dates can be wrong. Local-only processing is intended but not guaranteed by the catalog. I have not tested it on a NAS yet, so expert responses, model routing and saving behaviour still need checking there.
Built With
- markdown
- ocr
- sha-256
- speech-to-text
- ugos-skill
- ugreen-ai-nas
