Back to skill

Security audit

Hermes Agent Health Check

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Hermes Agent scanner, but it asks users to install and run an unpinned external Python package on local repositories.

Install only if you trust the `hermescheck` PyPI package and publisher. Prefer a pinned, hash-verified version in a virtual environment or container, run it only against the intended Hermes Agent checkout, avoid exposing unrelated credentials, and review generated reports before sharing them.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:23
Finding

Unpinned Third-Party Executable Dependency

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:23-27
Vulnerability Type: Unpinned third-party package installation
Risk Level: Medium

Vulnerable Code

bash
pip install hermescheck
hermescheck /path/to/hermes-agent

The same unsafe installation pattern also appears at README.md:82-90 and references/code-patterns.md:9-12.

Technical Analysis

The Skill directs users to install hermescheck from the default Python package index without specifying a version or verifying a cryptographic hash. The executable implementation is not included in this artifact, so its behavior cannot be validated by reviewing the Skill package.

Because the dependency is mutable, the code ultimately executed may differ from the version that existed when the Skill was reviewed. Compromise of the package publisher account, package-index distribution path, or a future release could introduce malicious behavior. Running a repository scanner necessarily gives the installed package read access to the target repository and the ambient privileges of the invoking user.

The installation is relevant to the declared auditing function, but using an unpinned and unverified executable dependency provides more supply-chain trust than is minimally necessary.

The static pre-scan’s other alerts were not confirmed as vulnerabilities:

  • README.md:31 is a GitHub Actions hyperlink with a Shields.io badge image; it does not download or execute a file.
  • Network and sensitive-data expressions in references/code-patterns.md are inert regular-expression examples used to describe scanner detections. They do not transmit data.
  • No embedded scripts, credential-harvesting logic, persistence installation, prompt hijacking, or active network-exfiltration implementation was found in the reviewed artifact.

Attack Path

  1. An attacker compromises the hermescheck package publisher account or causes a malicious future version to be distributed through the configured Python p ...[truncated 1218 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin hermescheck to a specifically reviewed version rather than installing the latest available release:

    bash
    python -m pip install "hermescheck==X.Y.Z"
    
  2. Publish and require cryptographic hashes through a locked requirements file:

    text
    hermescheck==X.Y.Z --hash=sha256:REVIEWED_PACKAGE_HASH
    

    Install it with:

    bash
    python -m pip install --require-hashes -r requirements-hermescheck.txt
    
  3. Record the reviewed package version, source repository commit, wheel hash, and expected publisher identity in the Skill documentation.

  4. Prefer a wheel built by a trusted, reproducible CI process. Avoid source-distribution installation where build hooks could execute unreviewed code.

  5. Run the scanner in an isolated environment with:

    • Read-only access limited to the intended target repository
    • Network access disabled during scanning
    • No inherited API keys or unrelated credentials
    • A dedicated unprivileged account or disposable container
    • Output write access limited to a designated report directory
  6. Update every occurrence of the installation command in SKILL.md, README.md, and references/code-patterns.md so users are not directed toward the unpinned alternative.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (6)

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 34)May include surrounding context.

md
<a href="https://github.com/huangrichao2020/hermescheck/actions/workflows/ci.yml"><img alt="CI" src="https://img.shields.io/github/actions/workflow/status/huangrichao2020/hermescheck/ci.yml?branch=main&label=CI&style=flat-square"></a>
  <a href="https://pypi.org/project/hermescheck/"><img alt="PyPI" src="https://img.shields.io/pypi/v/hermescheck?style=flat-square"></a>
  <a href="./LICENSE"><img alt="License" src="https://img.shields.io/github/license/huangrichao2020/hermescheck?style=flat-square"></a>
  <!-- ALL-CONTRIBUTORS-BADGE:START - Do not remove or modify this section -->
  <a href="#contributors"><img alt="All Contributors" src="https://img.shields.io/badge/all_contributors-1-orange.svg?style=flat-square"></a>
  <!-- ALL-CONTRIBUTORS-BADGE:END -->
</p>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 317)May include surrounding context.

md
real-world self-scan lessons.

<!-- ALL-CONTRIBUTORS-LIST:START - Do not remove or modify this section -->
<!-- prettier-ignore-start -->
<!-- markdownlint-disable -->
<table>
  <tbody>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 318)May include surrounding context.

md
<!-- ALL-CONTRIBUTORS-LIST:START - Do not remove or modify this section -->
<!-- prettier-ignore-start -->
<!-- markdownlint-disable -->
<table>
  <tbody>
    <tr>

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The enum for era_name permits only Chinese strings, which imposes a specific language on compliant outputs. In a general report schema, this is a natural-language locale restriction and no nearby description indicates user opt-in or a region-specific justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The document includes a dedicated 'Chinese' report example, but it does not explicitly state whether report language is user-selectable or optional. Because the policy requires avoiding forced language/locale behavior without opt-in, this wording could be interpreted as endorsing a fixed locale mode without clarifying user choice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

This is a markdown file, so missing-warning review applies. The instruction to run scans manually against a project path implies broad source inspection, but the document provides no warning that project contents may include secrets or sensitive data and should be reviewed carefully before sharing results.

Content

No source excerpt is available for this finding.