Back to skill

Security audit

Rtx Local Ai

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent local RTX/Ollama setup guide, with ordinary setup risks but no artifact-backed malicious or deceptive behavior.

Before installing, review the `ollama-herd` PyPI/GitHub project and consider using a virtual environment or pinned version. Be aware that prompts sent to the local router may be processed by machines in your configured RTX fleet, and the Ollama service/environment changes persist until you undo 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:31
Finding

Unpinned Third-Party Package Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:31
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

bash
pip install ollama-herd    # PyPI: https://pypi.org/project/ollama-herd/

Technical Analysis

The Skill directs users to install ollama-herd from PyPI without specifying an exact version or verifying an integrity hash. Consequently, the package resolved at installation time may differ from the version that existed when the Skill was reviewed.

The package provides the herd and herd-node executables used by the Skill, but their implementation is not included in this project. The audit therefore cannot verify the behavior of the code that users are instructed to install and run. If the package publisher, PyPI account, release process, or transitive dependency chain is compromised, an attacker could distribute modified executable code under this installation command.

Attack Path

  1. An attacker compromises the ollama-herd publishing account, release pipeline, source distribution, wheel, or a dependency resolved during installation.
  2. The attacker publishes a malicious release that remains compatible with the unpinned package name.
  3. A user follows the documented pip install ollama-herd instruction.
  4. pip resolves and installs the attacker-controlled release or dependency.
  5. Malicious behavior can occur during package installation if the selected package format invokes build code, or when the user subsequently executes herd or herd-node as instructed.
  6. The payload runs with the privileges of the user performing the installation or launching the installed commands.

Impact Assessment

Successful exploitation could permit arbitrary code execution under the installing or invoking user's account. The resulting code could access files, credentials, environment variables, local network services, GPU resources, and other ...[truncated 405 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin ollama-herd to an exact, reviewed version rather than installing the latest available release.
  2. Publish expected hashes for the approved wheel or source archive and require hash verification, such as through a locked requirements file used with pip install --require-hashes.
  3. Pin and audit all transitive dependencies using a reproducible lockfile.
  4. Prefer a prebuilt, reviewed wheel to avoid executing arbitrary source-build logic during installation.
  5. Install the package in an isolated virtual environment under an unprivileged user account.
  6. Document the trusted package source, release-signing or provenance-verification process, and the reviewed version.
  7. Re-audit the installed package and its dependencies before recommending any updated release.
Vulnerability Patterns
  • 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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 109)May include surrounding context.

bash
# RTX fleet overview
curl -s http://localhost:11435/fleet/status | python3 -m json.tool

# Check RTX GPU health
curl -s http://localhost:11435/dashboard/api/health | python3 -m json.tool

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 112)May include surrounding context.

md
curl -s http://localhost:11435/fleet/status | python3 -m json.tool

# Check RTX GPU health
curl -s http://localhost:11435/dashboard/api/health | python3 -m json.tool

# Models loaded on RTX GPUs
curl -s http://localhost:11435/api/ps | python3 -m json.tool

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 115)May include surrounding context.

curl -s http://localhost:11435/dashboard/api/health | python3 -m json.tool

Models loaded on RTX GPUs

curl -s http://localhost:11435/api/ps | python3 -m json.tool

text

Dashboard at `http://localhost:11435/dashboard` — live RTX performance monitoring.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown/manifest header provides a broad descriptive phrase for the skill but does not define any specific trigger phrases, activation boundaries, or exclusion conditions. In agent ecosystems that use descriptions for routing, generic language like 'turn your gaming PC into a local AI server' can match a wide range of AI or PC-related requests and lead to unintended invocation.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 85)May include surrounding context.

print(response.choices[0].message.content)

text

### curl

```bash
# RTX inference

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 133)May include surrounding context.

bash
# Linux
sudo systemctl edit ollama
# Add: Environment="OLLAMA_KEEP_ALIVE=-1"
# Add: Environment="OLLAMA_MAX_LOADED_MODELS=-1"
sudo systemctl restart ollama

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 136)May include surrounding context.

bash
# Linux
sudo systemctl edit ollama
# Add: Environment="OLLAMA_KEEP_ALIVE=-1"
# Add: Environment="OLLAMA_MAX_LOADED_MODELS=-1"
sudo systemctl restart ollama

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 143)May include surrounding context.

Image generation

bash
curl http://localhost:11435/api/generate-image \
  -d '{"model": "z-image-turbo", "prompt": "RTX-powered cyberpunk cityscape", "width": 1024, "height": 1024}'