---
type: "post"
id: "nous-research-released-hermes-security-an-official-hermes-d55f0e"
slug: "nous-research-released-hermes-security-an-official-hermes-d55f0e"
title: "Nous Research released hermes-security, an official Hermes Agent plugin for local repo security review."
description: "Nous Research’s official hermes-security plugin brings repository security review into Hermes Agent, with parallel reviewers, evidence-linked findings, optional test confirmation, and exportable reports."
retrieval_nugget: "Nous Research’s official hermes-security plugin brings repository security review into Hermes Agent, with parallel reviewers, evidence-linked findings, optional test confirmation, and exportable reports. For teams already using Hermes, the practical value is a structured review workflow; its limits matter just as much: model providers receive reviewed code, and test execution is not sandboxed."
published_at: "2026-11-08"
updated_at: "2026-11-08"
record_date: "2026-11-08"
date_kind: "published_at"
topics: ["hermes"]
entities: ["Nous Research","Hermes Agent","hermes-security","osv-scanner"]
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Nous Research added hermes-security, an official Hermes Agent plugin for local repository security reviews, with evidence-backed reports, multiple scan modes, and optional test confirmation."
prior_refs: ["https://newruntime.com/posts/nous-research-launches-hermes-marketplace-unifying-plugins-skills-cb1986/"]
source_url: "https://hermes-agent.nousresearch.com/marketplace/hermes-security"
source_urls: ["https://hermes-agent.nousresearch.com/marketplace/hermes-security"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:nous-research-released-hermes-security-an-official-hermes-d55f0e"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/nous-research-released-hermes-security-an-official-hermes-d55f0e/","markdown":"https://newruntime.com/posts/nous-research-released-hermes-security-an-official-hermes-d55f0e.md","json":"https://newruntime.com/posts/nous-research-released-hermes-security-an-official-hermes-d55f0e.json"}
markdown_url: "https://newruntime.com/posts/nous-research-released-hermes-security-an-official-hermes-d55f0e.md"
---

# Nous Research released hermes-security, an official Hermes Agent plugin for local repo security review.

## Retrieval answer

Nous Research’s official hermes-security plugin brings repository security review into Hermes Agent, with parallel reviewers, evidence-linked findings, optional test confirmation, and exportable reports. For teams already using Hermes, the practical value is a structured review workflow; its limits matter just as much: model providers receive reviewed code, and test execution is not sandboxed.

## Summary

Nous Research’s official `hermes-security` plugin brings repository security review into Hermes Agent, with parallel reviewers, evidence-linked findings, optional test confirmation, and exportable reports. For teams already using Hermes, the practical value is a structured review workflow; its limits matter just as much: model providers receive reviewed code, and test execution is not sandboxed.

## What happened

Nous Research describes `hermes-security` as a plugin for reviewing a local repository for security bugs. It requires Hermes 0.21.5 or later and Python 3.11 or later, and installs with `hermes plugins install hermes-security --enable`, followed by a restart. The plugin is Apache-2.0 licensed, with some parts adapted from Apache-2.0 sources. It offers standard, deep, and diff modes, a `/security` chat command, a `hermes security-review` CLI, and a Security page in Hermes Desktop. The CLI can also be run as `python -m hermes_security`.

A scan begins by mapping trust boundaries and attacker-reachable inputs, then dividing code into chunks for parallel subagents. The plugin says chunks are about 4,000 lines, with up to 10 reviewers running at once. A standard scan focuses on production code, including source, configuration, CI, infrastructure as code, and migrations; it normally skips tests, fixtures, documentation, examples, generated files, assets, and lockfiles, except for committed secrets. Skipped files and coverage gaps are recorded in the report. If production code fits within 90 chunks, about 360,000 lines, all of it is read; larger repositories are prioritized by risk-related code characteristics. Deep scans skip none of the standard scan’s file categories and make independent passes, three by default.

The report keeps evidence, rejected or inconclusive candidates, test results, and a record of code actually reviewed. It distinguishes code-backed findings from findings confirmed by tests, and exports to Markdown, JSON, CSV, and SARIF. An optional test-confirmation mode writes small reproductions and runs them in a temporary copy of the repository. Testing a running app requires a separate per-scan grant for one origin, one hour, and up to 200 requests; agent tools cannot issue that grant. Suggested patches are for review only—the plugin does not apply them.

Before review, a dependency check uses Google’s osv-scanner and checks declared runtime versions against endoflife.date. On the first scan it downloads a pinned osv-scanner release from GitHub, about 55 MB, and verifies its SHA-256. Package names and versions go to api.osv.dev and, for manifests without lockfiles, api.deps.dev; declared runtime versions go to endoflife.date. The check can be disabled with `HERMES_SECURITY_DEPENDENCY_CHECK=off`. Scan data and reports are stored in the active Hermes profile, but the configured model provider receives the code being reviewed. The listing identifies the plugin as version 0.1.1 and says it was updated two days before the page was captured.

## What the archive adds

New Runtime’s earlier coverage established Hermes Agent as a self-hosted, MIT-licensed open-source agent and tracked its move from v0.22.5 to v0.22.6, including migration support. [That coverage](https://newruntime.com/posts/nous-research-s-hermes-agent-mit-licensed-self-a50abd/) dates to 2026-10-24, and [the v0.22.6 note](https://newruntime.com/posts/nous-research-ships-hermes-agent-v0-22-6-4991af/) to 2026-11-03. On 2026-11-05, [New Runtime covered the Hermes Marketplace](https://newruntime.com/posts/nous-research-launches-hermes-marketplace-unifying-plugins-skills-cb1986/), which listed 976 plugins. This is a new capability within that ecosystem, not a restatement of the marketplace launch.

## Trend signal

This release supports the idea that agent security increasingly depends on explicit runtime boundaries: scan permissions, scoped app-testing grants, limits on automatic continuation, and recorded coverage make some actions and gaps visible. The most relevant pattern is [agent security moves to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/). The plugin also shows why controls need to be concrete: running repository tests still executes code as the user, so a temporary copy and credential-variable filtering are not a sandbox.

## Why it matters

For a team already running Hermes, the plugin offers a way to make AI-assisted security review more inspectable than a list of model-generated alerts. Review evidence, rejected candidates, unreviewed code, and test status travel with the report, which can help reviewers decide what to investigate rather than treating a clean-looking summary as proof. SARIF export also gives teams a format they can take into other security workflows.

The trade-off is that this remains a model-driven review, not an independent assurance mechanism. Teams must account for code being sent to their configured provider, inspect coverage gaps, and treat test-confirmed results carefully. Tests run with the user's privileges; the plugin itself warns that this is not a sandbox. Those boundaries make the tool most useful as an additional review layer with human ownership, not as a substitute for existing security testing or release gates.

## Limits and uncertainty

Nous Research says the model can miss issues or misread code, and that zero findings with partial coverage do not mean a repository is clean. The listing does not report independent evaluations of detection quality, false-positive rates, or performance across repository sizes. A completed scan may not continue through a Hermes restart, and deep scans have a fixed budget. Local storage does not mean local inference: the configured model provider receives code under review. Test confirmation runs repository code as the user, even in a temporary copy with credential-looking environment variables removed; that is explicitly not a sandbox.

## Skill to build now

If your team uses Hermes, try the plugin first on a repository you are authorized to share with your configured model provider. Start with a diff scan, then inspect the report’s coverage gaps and evidence before acting on any finding. Do not enable test confirmation on code you do not trust, and do not point it at a running app without reviewing the one-origin, one-hour, 200-request grant.

## Reader takeaway

`hermes-security` turns repository review into a resumable, evidence-bearing Hermes workflow, with parallel reviewers and portable reports. Its useful promise is traceability, not a guarantee that code is secure. Before adopting it, teams should decide whether sending source to their model provider is acceptable and keep test execution and scan coverage under human review.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Hermes](https://newruntime.com/topics/hermes.md): Explore the hermes topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)