Inspiration
Instrument loop documentation is spread across different files: datasheets, loop diagrams, calibration certificates, loop-check sheets and other project records. Before commissioning or handover, someone needs to establish whether the expected documents are present and whether important identifiers and revisions agree.
For smaller project teams, checking these records manually can be tedious and inconsistencies can be overlooked. I wanted to explore whether an AI NAS Skill could make this preliminary review more organised and repeatable.
What it does
LoopReady is an AI NAS Skill for instrument loop documentation inventory and conflict flagging.
A user provides a project folder, one or more instrument loop tags, and the document types expected for those loops. LoopReady is designed to discover relevant files, extract available document information, and organise the results into an inventory and a list of findings.
It distinguishes between missing documents, candidate duplicates, revision conflicts, tag mismatches, consistent fields and information that remains unverified.
The goal is to help a human reviewer identify where attention is needed, not to replace engineering judgement.
How we built it
LoopReady was designed as an orchestration-based Uliya Skill using UGREEN's documented filesystem, document-processing and knowledge-agent capabilities.
The intended workflow delegates file discovery to the filesystem agent and document analysis to the document agent. The parent Skill then organises the returned evidence into a report. An optional knowledge-agent step supports saving the report when requested and approved.
The Skill uses explicit input requirements, evidence-based comparison rules and completion/failure states. Its design also separates document findings from execution status, so incomplete evidence is not presented as a confirmed result.
OpenCode assisted with drafting and reviewing the Skill against the Uliya development kit.
Challenges we ran into
The main challenge was designing reliable comparisons without assuming that every difference is a conflict.
Different document types can have independent revision histories, and a document may legitimately reference several related instrument tags. LoopReady therefore uses narrower rules for identifying revision and tag conflicts.
Another challenge was the gap between documented agent capabilities and unverified runtime behaviour. In particular, the development kit does not establish exactly how discovered file paths are handed between agents. We have kept this limitation explicit rather than claiming an end-to-end workflow that has not been tested.
Accomplishments that we're proud of
We developed a focused Skill specification with a defined input workflow, three registered agent delegations, explicit finding categories and safeguards against unsupported engineering conclusions.
We are particularly proud of the distinction between a confirmed discrepancy and an item that simply needs further review. In engineering documentation, that distinction matters.
What we learned
Building an AI Skill requires more than connecting agents. It also requires understanding their documented capabilities, defining what evidence is sufficient for a conclusion, handling incomplete results and communicating uncertainty.
We also learned that a technically plausible workflow should not be presented as a verified one until it has been tested in its target environment.
What's next for LoopReady
The next stage is testing on a UGREEN AI NAS. This will focus on filesystem discovery, document extraction, path handoff between agents, revision comparison and optional report saving.
Further work will explore how reliably LoopReady handles real-world instrument documentation, including inconsistent naming, incomplete records and scanned documents.
The long-term goal is a lightweight pre-ingestion documentation review tool that helps project teams prepare more organised records for formal engineering and commissioning workflows.
Built With
- automation
- document
- language
- natural
- python
