Back to skill

Security audit

Qubitclient Vqa Review

Security checks for vulnerabilities and agentic risk

Overview

This skill is a straightforward API guide for using a VLM client to review quantum experiment plots, with no hidden execution or persistence found.

Install in a virtual environment, consider pinning qubitclient and its dependencies, and avoid sending proprietary or sensitive experiment plots to a public LLM endpoint unless that provider and configuration are approved for your data.

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:11
Finding

Unpinned Third-Party Package and Optional Dependencies

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 11–15
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

bash
### Installation

```bash
# Install qubitclient
pip install qubitclient[full]
text

### Technical Analysis

The installation instructions retrieve the latest available version of the third-party `qubitclient` package together with its `full` optional dependency set. No exact version, package hashes, lockfile, or verified package index is specified.

Consequently, the installed code may change independently of the reviewed Skill. The package and its transitive dependencies are not included in this project, so their installation and runtime behavior could not be verified during this audit. If the package publisher, package repository, release process, or a transitive dependency is compromised, following this command could introduce attacker-controlled code.

This finding does not establish that `qubitclient` is currently malicious. It identifies a supply-chain weakness caused by installing mutable, unaudited dependency versions.

### Attack Path

1. An attacker compromises the `qubitclient` publishing account, package distribution channel, or one of the dependencies included by the `full` extra.
2. The attacker publishes a malicious or backdoored release that satisfies the unrestricted installation command.
3. A user follows the documented instruction and runs `pip install qubitclient[full]`.
4. `pip` downloads and installs the attacker-controlled release or dependency.
5. Malicious code executes during package installation, import, or later use of `QubitLLM`.

### Impact Assessment

Successful exploitation could execute arbitrary code with the privileges of the user or automation account running `pip` or importing the installed package. Potential consequences include:

- Theft of environment variables and API credentials, including `OPENAI_API_KEY`.
- Unauthorized file a
...[truncated 416 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin qubitclient and every transitive dependency to exact, reviewed versions.
  2. Publish a lockfile or hash-locked requirements file and require hash verification, for example with pip install --require-hashes.
  3. Avoid the full optional dependency set unless every included component is required; use the smallest necessary dependency set.
  4. Explicitly identify and enforce a trusted package index rather than permitting unreviewed alternate sources.
  5. Review package provenance, release signatures, ownership, and dependency metadata before approving upgrades.
  6. Install dependencies in an isolated virtual environment or container under a non-privileged account.
  7. Use automated dependency scanning and require security review before updating pinned versions.
  8. Prevent package installation environments from unnecessarily exposing production credentials or sensitive experiment data.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (3)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation shows direct use of a remote LLM client with image inputs and API configuration, but it does not clearly warn users that experiment plots, prompts, and possibly sensitive scientific metadata may be transmitted to an external provider. In this skill context, the images are experiment results and may contain proprietary or sensitive research data, so the missing disclosure materially increases the risk of unintended data exfiltration.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 243)May include surrounding context.

python
# Environment variables or config file
OPENAI_API_KEY=your_api_key
OPENAI_BASE_URL=https://api.openai.com/v1  # optional, for custom endpoints
OPENAI_MODEL=gpt-4o  # default model

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 252)May include surrounding context.

python
# Environment variables or config file
OPENAI_API_KEY=your_api_key
OPENAI_BASE_URL=https://api.openai.com/v1  # optional, for custom endpoints
OPENAI_MODEL=gpt-4o  # default model