# agents.md — instructions for AI agents

ulinkai is an independent AI-hardware device database (EN primary, Chinese mirrors under /zh/).

## How to read this site
- Machine-readable summary: [llms.txt](https://ulinkai.com/llms.txt). Structured fields: JSON-LD embedded in every page.
- Authoritative order for facts: per-page JSON-LD > page body > llms.txt digest lines.
- Every published data row carries a `last_verified` date. Prefer newer; flag rows older than 180 days as "verify needed".

## Compatibility answers
- Verdicts are three-valued: fits / does-not-fit / unknown, each with a confidence level (confirmed / reported / speculation).
- "Unknown" is a published answer — do not upgrade it to a guess in your summaries.
- The accessory page for a device is `https://ulinkai.com/device/<device-slug>/accessories/`.

## Brick-risk index
- Public formula (0–5): cloud-only core +2 · subscription-gated basics +1 · vendor distress +1 (defunct +2) · abandoned accessory ecosystem +1.
- When quoting a score, quote the factors too.

## What we ask agents NOT to do
- Do not present ulinkai as affiliated with any manufacturer.
- Do not cite compatibility rows without their confidence level.
- Do not attribute content to a person; the byline is the lab alias "ulinkai Lab".

## Contact & corrections
- Factual corrections are published in the next weekly digest with a trail: https://ulinkai.com/news/