case-file-evidence-indexer

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 case-file-evidence-indexer

What it does

Indexes a case folder on the NAS into a source-cited research brief with a timeline, people list and claims register marking each claim supported, conflicting or unsupported.

Target user and problem

Documentary researchers, true-crime and history writers, and students working from case files on a NAS. Facts in large mixed folders are hard to trace to sources, so unsourced claims slip into scripts, and sensitive material should not go to cloud tools.

Proposed solution

NAS experts read the folder, transcribe recordings, and extract dated events and named people with verbatim quotes and locations. Each claim, including those in the user's own draft script, is checked against that folder only. Gaps and unsupported claims come first in the brief.

Core user flow

User names a case folder (optionally a draft script); read-only inventory; documents read, recordings transcribed, scans read as text; timeline, people and claims drafted with citations; claims verified in-folder, conflicts cross-checked; draft shown in chat; on confirmation, saved as a new file, never overwriting.

What is different

Conventional summarisers answer from model memory and blend sources. This Skill allows only passages the experts return, cites file and page or time on every line, reports missing support as "unsupported" rather than filling gaps, checks a writer's own draft, and keeps data on the NAS.

UGREEN AI NAS capabilities used

Inventory: ug-filesystem (find, ls, stat), ug-filemgr (mediainfo). Documents: ug-document (metadata, extract, qa). Recordings: ug-knowledge (audio-transcription). Scans: ug-photo (vision-ocr). Verification: ug-knowledge (collect-evidence), ug-document (reconcile). Saving, after confirmation: ug-filesystem (mkdir), ug-knowledge (save-text).

Technical approach

An orchestration-only SKILL.md following the Uliya standard. The parent delegates to file-agent, document-agent, knowledge-agent and photo-agent, passing only real results between steps; no scripts. Inputs: case folder, optional draft script. Permissions: read-only except two confirmed writes (new folder, new file). Models: the NAS experts' own; no external service. Content stays on the NAS.

Development status

Implemented: SKILL.md and five reference files, passing the UGOS Skill Creation Tool static checker with no diagnostics. Pending NAS validation if selected as Finalist: installation, any NAS run, delegation wording, citation accuracy, recording support. AI disclosure: An AI assistant was used as a drafting and editing tool for the skill files and write-up; the concept, direction and review were by the entrant.

Validation plan

Core 1: three documents and one recording yield a timeline citing a real file and page or time offset on every row. Core 2: a draft script with five statements (two unsourced, one date conflict) yields three supported, two unsupported, and the conflict shown with both sides. Edge: an unreadable file gives a partial brief listing it; an empty folder reports no files and stops.

Known limitations

Untested on a NAS. Transcription, extraction and scan quality depend on the NAS experts; handwriting and poor scans may fail. Video gets technical details and audio transcription only, with no visual analysis. "Unsupported" means not found in the folder, not false. No dependencies beyond the NAS experts.

Third-party components

UGOS Skill Creation Tool (uliya-skill-kit 0.7.3), from the UGREEN Hackathon Starter Kit (ai.ugreen.com/pages/hackathon-starter-kit); not bundled in the ZIP. UGREEN AI NAS registered experts (file-agent, document-agent, knowledge-agent, photo-agent): run-time only, nothing bundled. AI drafting assistant: used to draft and edit the Markdown text of the package and write-up only. External services at run time: none.

Built With

• ai
• markdown
• nas
• ugos
• ugreen