Inspiration

Research folders contain papers, drafts, experimental notes, and simulations with similar names but different assumptions. LabTrace makes the evidence behind a conclusion inspectable before it is copied into a report.

What it does

Select a NAS folder and ask a research question. LabTrace inventories a bounded batch, requests source-grounded answers, and produces an evidence table with actual source locators, units, methods, and conditions. It preserves experiment versus simulation, identifies incomparable conditions and missing evidence, and rechecks available source fingerprints. Saving a reviewed brief to a new file is optional and requires platform approval.

How it is built

This orchestration-only Skill was authored using the official UGOS Skill Creation Tool 0.7.3. It delegates supported operations to file-agent, document-agent, and knowledge-agent. No executable runtime, custom network integration, or direct NAS access is bundled. Its evidence contract distinguishes conditional support, comparable contradiction, condition mismatch, evidence not found, and insufficient extraction.

Development status and validation

Round 1 pre-test Skill: the official local checker reports static-pass, zero errors and zero warnings. NAS installation, expert execution, source locator accuracy, model locality, and verified report saving remain untested. Four original synthetic fixture PDFs and a manual validation plan are prepared; these are not scientific results or a NAS demonstration.

Challenges and next steps

The main design challenge is avoiding confident comparisons when conditions differ or extraction is incomplete. Planned target-device tests cover source-grounded auditing, preview-only behavior, approved non-overwriting export, unavailable evidence, and hostile text embedded in documents.

AI and dependency disclosure

OpenAI Codex proposed the name and direction, drafted the Skill, proposal and submission text, and operated development tools. The official UGOS tool and Uliya capabilities are subject to UGREEN agreements. No third-party runtime library, model or dataset is bundled. Local PDF preparation uses ReportLab (BSD) and pypdf (BSD-3-Clause), outside the Skill ZIP.

Built With

Share this project:

Updates

Submission history