Inspiration
A green status is not the same thing as proof.
In operational work, I've seen how easily a dashboard, workflow or AI assistant can report that something is complete while the actual evidence is missing. A backup may be stale, an export may never have reached its destination, or a receipt may not exist.
I wanted to build something that asks a better question: What evidence makes this claim true?
That became Proof Before Ready.
What it does
Proof Before Ready is an evidence-first Skill designed for UGREEN AI NAS. Given a completion claim and an authorized folder, it searches for supporting files, checks available metadata and optionally compares SHA-256 hashes.
It returns one of three outcomes:
- Ready: The specified file evidence meets the checks.
- Not ready yet: Required evidence is missing, stale or mismatched.
- Unable to verify: Permissions, ambiguity or another limitation prevent a reliable conclusion.
Every result identifies the evidence examined and the first unmet requirement.
How I built it
I developed a Pre-test Skill Package using Markdown instructions and reference files, structured around Uliya's expert-agent model and the documented UGOS 0.7.7 action catalog.
The parent Skill delegates tasks to specialized agents. File discovery and metadata checks belong to the file agent; document extraction is used only when needed; saving an evidence card requires user approval.
The workflow is deliberately permission-scoped. Missing evidence never authorizes a broader search or fabricated substitute.
Challenges and lessons learned
The biggest challenge was defining what Ready can legitimately mean.
A file's existence does not necessarily prove the underlying business outcome. A matching hash establishes file identity, not approval. And missing evidence is different from evidence that cannot be accessed.
Those distinctions shaped the three-outcome design and the requirement to stop when verification cannot continue safely.
Current status and what's next
The Pre-test Skill Package is built, and a local static checker pass has been reported. Real NAS execution remains untested and is planned for the finalist validation stage, if selected.
The next tests will examine matching backups, missing files, stale evidence, hash mismatches and denied folder access.
My goal is simple: build AI workflows that earn their completion status through evidence instead of confidence alone.
AI assistance was used during development and documentation, with human direction and review. The Skill's reference package includes further disclosure information.
Built With
- agent
- ai
- automation
- markdown
- nas
- ugreen
- uliya
- workflow
