Inspiration
A delivery notification says your parcel has arrived, but there's nothing at your door. Even with security cameras, finding the relevant moment can mean searching through dozens of detections across multiple feeds.
Worse, an empty search result can be mistaken for proof that nothing happened.
We wanted to make the first step of investigating a missing delivery faster, clearer and more trustworthy.
What it does
Delivery Window Check is an AI-assisted UGREEN NAS skill designed to help users identify which camera events deserve attention after a reported parcel delivery.
The user supplies a reported delivery time, selects up to two authorised cameras and chooses a review window.
The skill is designed to:
- Search recorded package-detection events on each camera.
- Rank relevant detections by proximity to the reported delivery time.
- Suggest which camera and timestamp to inspect first.
- Distinguish between no detected events, incomplete searches and unavailable results.
- Let users select up to three events for playback-position lookup.
Importantly, it never claims that a detection proves a parcel was delivered.
How we built it
The skill was authored using OpenAI Codex and UGREEN's official Uliya Skill Creation Kit.
It uses the registered security-monitor-agent and documented ug-surveillance capabilities to define a read-only workflow around native NAS camera history.
The design prioritises bounded searches, explicit uncertainty, authorised access and local processing.
Challenges and development
One of the main design challenges was handling incomplete evidence responsibly.
A camera returning zero detections is not the same as a camera failing to return results. Similarly, an available playback position does not establish what occurred in the footage.
We designed the workflow to preserve these distinctions instead of generating misleading conclusions.
The official local static checker passed with zero errors and warnings.
Current limitation: This is a Round 1 skill specification. Actual NAS installation, camera-event responses and playback integration remain untested.
What's next
If selected as a finalist, we'll use the provided NAS testing environment to validate event retrieval, camera-specific results, ranking accuracy, playback references and edge cases.
The goal is a useful, privacy-conscious assistant that helps people find the right evidence to review without pretending to know more than the camera records can establish.
Built With
- markdown
- openai-codex
- ug-surveillance
- ugreen-ugos
- uliya-skill-creation-kit
