Inspiration

In project management, a single approved decision can affect dozens of documents: requirements, procedures, schedules, reports, and technical specifications.

The challenge is not simply finding information. It is understanding what changed, which documents are affected, which versions are valid, and what must remain untouched.

We envisioned DossierTrace as an AI-powered assistant that helps project managers maintain consistency across their documentation, using UGREEN AI NAS as the environment where their files already live.

What it does

DossierTrace turns project decisions into evidence-backed document impact reviews.

When a project manager selects a decision and authorized NAS folders, the Skill is designed to:

  • Identify relevant documents and examine available evidence of approval and version status.
  • Detect inconsistencies between approved decisions and existing documentation.
  • Identify dependent processes that must remain unchanged.
  • Generate an executive summary explaining the change, its impacts, and supporting evidence.
  • Prepare proposed updates for individual documents.
  • Highlight potential schedule impacts without automatically changing deadlines.
  • Preserve original documents and historical records.

Rather than simply summarizing files, DossierTrace focuses on the relationships between decisions, documents, and project consequences.

How we built it

We designed DossierTrace around the official UGREEN Uliya Skill architecture.

The Round 1 pre-test package includes a newly authored SKILL.md and seven linked Markdown references defining the workflow, decision rules, evidence requirements, and failure handling.

Its architecture uses documented NAS capabilities, including ug-filesystem for file discovery and ug-document for extraction and comparison, orchestrated through registered expert agents.

We also designed human approval checkpoints, version identification rules, impact classifications, and safeguards against unauthorized modifications.

ChatGPT assisted with architecture exploration, technical documentation, and preparation of the Skill. Development and technical planning were managed using GitHub-based project documentation.

Challenges we ran into

Our biggest challenge was designing a reliable document-change workflow without assuming capabilities that had not been verified on a physical NAS.

Several issues required particular attention:

  • Distinguishing an approved document from a newer, unapproved draft.
  • Identifying indirect dependencies without generating false impacts.
  • Handling Word, Excel, PDF, Microsoft Project, Markdown, and CSV consistently.
  • Preserving historical documents and preventing unauthorized changes.
  • Differentiating a proposed update from a successfully applied one.
  • Handling incomplete information, conflicting approvals, and uncertain user identity.

We addressed these through explicit evidence requirements, conservative operation controls, and a clear separation between analysis, human approval, and execution.

Accomplishments that we're proud of

We developed a structured product design covering the complete document-change review lifecycle.

Our work includes:

  • A defined architecture for evidence collection, version assessment, impact analysis, human review, and traceability.
  • 31 documented development tasks with acceptance criteria.
  • 21 defined contract and validation test cases.
  • A newly authored Round 1 Skill package with linked technical references.
  • Explicit protections for original documents and unaffected workflows.
  • A planned fallback that provides Markdown update proposals when native editing cannot be safely supported.

The package has undergone local structural checks. Full execution and validation on UGREEN AI NAS remain pending.

What we learned

We learned that reliable AI-assisted document management requires more than extracting text or generating summaries.

A document's latest modification date does not establish its authority. A business decision does not automatically authorize file changes. And a successful tool response does not necessarily prove that the intended result was safely saved.

These distinctions shaped DossierTrace into an evidence-first workflow where uncertainty is visible and human decisions remain central.

We also learned the importance of designing around documented platform capabilities rather than assuming that every NAS feature is accessible to an AI Skill.

What's next for DossierTrace — Project Change Review

Our next objective is to validate the Skill against the current official UGREEN development kit and, if selected as finalists, test it on the assigned UGREEN AI NAS.

The planned validation includes document discovery, evidence extraction, version identification, impact accuracy, permission handling, and safe operation controls.

We also intend to evaluate representative multi-document scenarios, including conflicting approvals, shared workflows, and project schedules.

Our long-term vision is to help project managers move from manually checking disconnected documents to reviewing a clear, traceable picture of how each approved decision affects their project.

Built With

  • document-agent
  • file-agent
  • github
  • markdown
  • ug-document
  • ug-filesystem
  • ugos
  • ugreen-ai-nas
  • uliya
Share this project:

Updates

Submission history