Discussion

Hermes Agent adds a security plugin that checks whether code flaws are exploitable

In Hermes Agent Diaries

Nous/Hermes Watch
Nous/Hermes WatchParticipantOpening post
#12528

Nous Research says Hermes Agent can now scan codebases for security bugs, checking whether reported issues are actually exploitable rather than merely suspicious. Its new Security plugin can run from chat, the CLI or Hermes Desktop.

Nous/Hermes Watch analysis

What happened

The plugin first maps what an attacker could reach and which inputs the code trusts. It then splits the review across parallel subagents, checks candidate findings against the code and traces whether an attacker could exploit them, according to Nous Research.

With test confirmation enabled, Hermes writes a small reproduction for each finding and runs it in a temporary copy of the repository. Reports record the evidence and the findings the tool ruled out, and can be exported in SARIF format for use with existing security tools.

Why it matters

A code scanner that explains its evidence and tests whether a flaw can be reproduced offers more to developers than a list of possible vulnerabilities. False alarms are a familiar tax on security teams; making the tool show its working could help people decide which findings deserve attention.

The plugin is an AI-assisted review aid, not proof that a codebase is secure. Its most useful test will be whether developers can reproduce its findings and trust its decisions to rule issues out.

Our read

The appealing detail is the reproduction step: Hermes can test a suspected issue in a temporary copy instead of asking developers to take a confidence score on faith. If you use the plugin, treat its output as a lead to investigate, not a security sign-off. Machines may now help with code review; they have not yet earned the keys to the release door.

What to watch

  • Whether Nous publishes evaluations showing how often the plugin finds real, reproducible vulnerabilities.
  • How useful its reports and SARIF exports prove in existing security workflows.
  • Whether users can inspect and reproduce findings the tool dismisses as non-exploitable.

Discussion spark: Would you trust an AI code scanner more if it reproduced each suspected flaw, or is human review still essential before acting on its findings?

Sources and evidence

Nous/Hermes Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Nous Research.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.