ByteDance Agent TARS / UI-TARS is listed on Protagentic — may we show its rating?

[Protagentic]

Hello,

I'm writing on behalf of Protagentic (https://protagentic.ai), an independent index of AI agents. ByteDance Agent TARS / UI-TARS (bytedance/UI-TARS-desktop) is listed there, along with about 180 others, because it's one of the projects people actually use.

We rate the projects we list on published, evidence-based criteria — every number comes from something we can point to, and the method is public at https://protagentic.ai/methodology. We have computed a rating for ByteDance Agent TARS / UI-TARS. We do not show anyone's rating without their agreement, which is why I'm writing.

You can see the rating, and every measurement behind it, before deciding: claim the listing at https://protagentic.ai/agents/db3dd9d4-eb93-4178-d55c-c39c69d24cd8 — it takes a small file in the repository and no account details from us — and it is shown to you and nobody else.

If you're happy for it to be public, reply "yes" here or email admin@protagentic.ai, or switch it on yourself from the listing once claimed. You can contest any measurement, and withdraw at any time — withdrawing hides the grade but keeps the history, so nothing is lost.

If you'd rather not be shown, a one-line reply is enough and we won't ask again.

Thank you for the work on ByteDance Agent TARS / UI-TARS.

— Sabin Ewing, Protagentic (admin@protagentic.ai)

[zzzz0902zzzz-rgb]

Not a maintainer of bytedance/UI-TARS-desktop, so nothing here is consent to publish — that call is theirs. But if a claim file is going to be added, the verification path is worth pinning down first:

• Compare the fetched file's exact content (or a normalized hash), never a substring test. token in text comes back true when the token also appears in a comment, a changelog, or the listing page itself → false green on ownership.
• Write the raw response of each attempt to its own file (status line + body). If a retry overwrites the same file, you've destroyed the evidence of which attempt actually succeeded.
• On failure, print the reason. A bare "no"/"not verified" is indistinguishable from a network error, and a silent fallback — fetch fails but the page still renders a "claimed" badge from a cached copy — looks like a verified claim while nothing was proven.

So for whoever decides: fetch the raw file URL, print the status code plus the first bytes of the body, and require the badge to flip only on an exact-match success. If the methodology page doesn't state how the claim file is validated, ask before adding it.