Back to skill

Security audit

Private Ai

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent local private-AI setup guide, with the main caution being normal supply-chain risk from installing and running an unpinned third-party Python package.

Install `ollama-herd` only from the expected publisher/source, preferably in a virtual environment or other isolated non-admin context. Review or pin the package version before sending sensitive documents, audio, or prompts through the local service, and confirm your local-network exposure matches your privacy needs.

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:25
Finding

Unpinned Third-Party Package Is Installed and Executed

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 25-27
Vulnerability Type: Unpinned executable dependency
Risk Level: Medium

bash
pip install ollama-herd    # install private AI router
herd                       # start private AI router (port 11435)
herd-node                  # run on each device — private AI nodes discover each other

Technical Analysis

The setup instructions install ollama-herd from the package index without pinning an exact version or verifying an integrity hash. They then instruct the user to execute the package-provided herd and herd-node commands.

Python packages can execute installation or runtime code with the privileges of the invoking user. Because the dependency's source code is not included in the audited project, its networking, telemetry, file-access, and model-download behavior cannot be verified from this artifact. An unrestricted installation may also resolve to a future package release that differs from the version originally reviewed.

Installing a third-party component is necessary for the documented functionality, but accepting any available version without cryptographic verification exceeds the minimum supply-chain trust needed for a reproducible installation.

Attack Path

  1. An attacker compromises the package maintainer account, package repository, release pipeline, or another component in the dependency chain.
  2. The attacker publishes a malicious release under the expected ollama-herd package name.
  3. A user follows the documented pip install ollama-herd instruction, and pip resolves the malicious release because no version or hash is specified.
  4. Malicious code may run during installation or when the user invokes herd or herd-node.
  5. The payload executes with the user's privileges and can access resources available to that account.

This is a conditional supply-chain exploitation path; the audit found no evidence that the c ...[truncated 669 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin ollama-herd to a specific, reviewed version rather than allowing unrestricted package resolution.
  2. Publish a lock file or requirements file containing cryptographic hashes and instruct users to install with hash enforcement, such as pip install --require-hashes -r requirements.txt.
  3. Document the authoritative package index and upstream source repository so users can verify package provenance.
  4. Sign releases or publish verifiable attestations and checksums for distributed artifacts.
  5. Recommend installation in a dedicated virtual environment or other isolated, non-administrative environment.
  6. Review the pinned package and its transitive dependencies for installation hooks, external network access, telemetry, sensitive-file access, and unsafe command execution before recommending it.
  7. Avoid instructing users to run installation or service commands with elevated privileges unless a narrowly scoped privilege is demonstrably required.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

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 118)May include surrounding context.

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

# Private AI health checks
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 121)May include surrounding context.

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

Private AI health checks

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

text

Private AI dashboard at `http://localhost:11435/dashboard` — accessible only on your local network.

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 54)May include surrounding context.

bash
# Private AI with DeepSeek-R1 — no data sent to cloud
curl http://localhost:11435/api/chat -d '{
  "model": "deepseek-r1:70b",
  "messages": [{"role": "user", "content": "Analyze this private financial report with offline AI"}],
  "stream": false