Skip to content
DeepSeekDetect
EN

Can you detect text written by DeepSeek's Super-agent?

As of 2026-11-12, text from DeepSeek's Super-agent can be checked like any other prose, but the result cannot tell you which model wrote it. The prototype's setup guide leaves the model to the operator, so a pull request description or reply it posts may come from any configured provider. A paste of at least 100 words shows whether it reads as AI-written and how close the style is to DeepSeek and other families. That is a style signal, not proof.

What DeepSeek released

Bloomberg reported on 2026-11-01 that DeepSeek had released a prototype of a more autonomous coding agent. The code is public on GitHub as dsh-superagent-prototype under the MIT licence. Its README calls it "a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness". When an allowed person asks for work, it opens a work item, works on it in its own session and "answers in the original conversation, for example with a pull request". It listens to GitHub and, optionally, to Feishu chats.

The README is plain about status: "Research prototype for reference only, not an official DeepSeek product" and "Treat it as research code, not production-ready software." DeepSeek Harness, the platform it is packaged for, is also MIT-licensed and in developer preview, with a warning that there will be compatibility-breaking changes.

What is confirmed

Point Status Source
Open-source research prototype, MIT licence Confirmed Repository README
Replies and opens pull requests in the conversation it was asked in Confirmed Same README
Writes under a dedicated GitHub account Recommended by the setup guide: "everything the agent writes appears under that name" SETUP.md
Which model writes the text Not stated; the operator configures a model provider SETUP.md
Labels or discloses AI-authored output The README does not mention it Same README
Release timing Reported by Bloomberg on 2026-11-01; the repository page shows no release date Bloomberg, 2026-11-01

Why the model question matters

A Veread result has two parts: whether the text reads as AI-written, and how close its style is to each model family. Because the operator picks the provider, text from the agent may look like DeepSeek, or like a different family. A low DeepSeek similarity therefore says nothing about whether the agent wrote it, so read the AI verdict first and the family list second.

The published figures, on the fact sheet, are the ones that apply here. On AI models never used in development Veread caught 94.9%, which matters when an operator can plug in any provider. For fully AI-written texts from the DeepSeek family the figure is 95.1%.

What a text check can do for a maintainer

For the question "is this account a bot", account and repository history answer more directly than style, and the setup guide itself recommends a dedicated account for the agent. GitHub's own thinking points the same way. The Register reported on 2026-02-03 that GitHub was weighing attribution mechanisms that flag when AI tools were used, and product manager Matthew Isabel said, "A bad or off-topic PR is a bad PR, regardless of where it came from."

Labels and watermarks

DeepSeek's documented labels are notices inside its own platform, and no hidden text watermark or public checker has been published; the watermark status page has the dated details. Nothing verified ties those labels to text a self-hosted agent posts to GitHub. Phrase habits that separate DeepSeek from human writing are listed in how to tell if DeepSeek wrote it.

Check a text

Paste the text into the DeepSeek AI detector and run the check. The result is writing-style similarity, not proof of authorship, and the minimum is 100 words. Teams that want checks inside a review workflow can use Fileverity's API.

Verified 2026-11-12 against the GitHub repositories and the sources linked above.

Questions and answers

Can a detector tell that DeepSeek's Super-agent wrote a pull request description?

Not by itself. A detector reports whether a text reads as AI-written and how close its style is to DeepSeek and other families, but the prototype leaves the choice of model to the operator, and a style result is never proof of who or what wrote the text.

Is the Super-agent an official DeepSeek product?

No. The README of the dsh-superagent-prototype repository calls it a "Research prototype for reference only, not an official DeepSeek product".

Which model does the Super-agent use?

Its published setup guide does not name one. It says DeepSeek Harness needs a model provider and that configuring one is an operator step.

Does the text the agent writes carry a label?

The README does not mention labelling or disclosure of what the agent writes. DeepSeek's documented labels are notices inside its own platform.

How much text do I need to check?

At least 100 words. A long pull request description or review comment can be checked; a one-line reply cannot.