What it does
SearchCanary is designed for NAS owners who want to know whether a photo search they rely on still finds the same important files. It remembers a search that worked and checks it again when the owner asks.
Illustrative example - not an observed NAS test: Yesterday, "Dad's birthday cake" found an important photo. Later, SearchCanary checks whether the same search still finds that exact photo.
How the owner uses it
Say "Remember this search" and choose up to three important photos from its results. SearchCanary verifies the selected files and repeats the same search. It then offers to save an approved baseline: a record of the exact search, files and result for later checks.
Later, ask for another check. Each photo gets a separate result:
- PASS: the verified expected photo appears. A verified hit can pass even if other results are hidden; that coverage limit stays visible.
- MISS: the search is complete and the same accessible, unchanged expected photo is absent.
- UNKNOWN: incomplete results, uncertain file identity, changed files or other missing evidence prevent a reliable conclusion.
Changed search conditions can prevent comparison with the baseline. SearchCanary keeps the current result separate from whether a reliable comparison is possible.
For one affected photo, the owner may approve up to two alternative searches. Recommend an alternative only if it returns the verified correct file. Show the original result and the alternative separately. A successful alternative does not mean the original search was repaired.
Why this is useful
Ordinary photo search helps find a photo now. SearchCanary checks whether a previously successful search keeps finding the owner's selected files. An accessible file can be absent from search results; a missing result alone does not mean the photo was deleted.
How it is designed to work
The Markdown Uliya Skill coordinates UGREEN's registered Photo, File and Knowledge experts. The Photo expert searches. The File expert checks the exact paths and SHA-256 fingerprints, which detect changed file contents. The Knowledge expert saves an approved Markdown report. The File expert reopens that report directly for later checks.
The workflow requires owner-confirmed Local model mode before handling private data. It adds no external service or custom runtime. Local confirmation is a workflow requirement, not independently verified network isolation. Original photos remain unchanged. Every report save needs approval of the exact content and destination, plus the platform's approval flow. Denied saving does not erase the in-chat results.
What is prepared today
Version 1.2 Skill instructions, the baseline format and acceptance scenarios are written. The official UGREEN toolkit 0.7.7 static checker was rerun on the final package on November 14, 2026: zero errors and zero warnings. Static validation checks the package; it does not prove installation or runtime behavior.
OpenAI Codex assisted research, drafting, workflow specification, package preparation, PDF creation and review. The Skill has not run on a UGREEN NAS. Installation, supported scope resolution, returned file identities, complete-result handling, alternative searches and baseline saving/reopening still need NAS testing.
Limits and next steps
Content search supports personal/shared scope but has no documented folder filter. Folder-only checks require supported structured filters. A partial preview cannot prove a photo is absent. Changed files or conditions can prevent comparison, and alternatives may fail.
These checks cover selected examples. They do not measure the whole library, prove an indexing defect, guarantee finding a photo or repair the search engine.
If selected as a finalist, first verify search scope and file identity on the assigned NAS. Then test successful capture and repeat checks, complete misses when observable, partial results, denied saving, saved-report reopening and the bounded alternative-search flow. Label any simulated miss as a simulation.
Built With
- markdown
- ugos
- ugreen
- uliya
