Back to skill

Security audit

Local GMNCODE Vision

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent image-analysis purpose, but it tells the agent to run an unbundled local script that sends images to an external API using a local API key.

Review before installing. This is not evidence of malware, but only install it if you trust and can inspect the local script at /home/ubuntu/.openclaw/workspace/scripts_gmncode_image.py, understand that images and prompts may be sent to GMNCODE, and are comfortable exposing GMNCODE_API_KEY to that script's process.

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:17
Finding

Unverifiable External Script Executed from a Mutable Absolute Path

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:17-24
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

The skill depends on an executable located outside the audited project:

text
/home/ubuntu/.openclaw/workspace/scripts_gmncode_image.py

It directs the agent to invoke that external dependency as follows:

bash
/home/ubuntu/.openclaw/workspace/scripts_gmncode_image.py <image_path> "your prompt"

The same dependency is also documented in README.md:5-12. The project contains only SKILL.md and README.md; therefore, the referenced script, its network endpoint, its dependency chain, and its handling of images and credentials could not be audited.

Technical Analysis

The skill executes an absolute-path script that is not bundled, version-pinned, or integrity-verified by the audited package. The documented script requires GMNCODE_API_KEY and is intended to transmit image content to an external Responses API.

Because the script exists outside the skill package, its behavior can change independently after this skill has been reviewed. If another user, package, process, or compromised update mechanism can modify the referenced file, legitimate-looking invocations of the skill will execute substituted code.

A substituted script would run with the permissions of the agent process. It could read the supplied image, inspect environment variables inherited by the process, access other files permitted to that account, invoke local commands, or communicate with unauthorized network destinations. The audit found no evidence that the current external script is malicious; the risk arises because its implementation and integrity are outside the reviewed artifact.

Attack Path

  1. An attacker obtains write access to /home/ubuntu/.openclaw/workspace/scripts_gmncode_image.py or controls the mechanism that installs or updates it.
  2. The attacker replaces or modifies the scr ...[truncated 1390 chars]
Remediation
View remediation

Remediation Suggestions

  1. Bundle the image-processing script inside the skill package so its implementation is included in security reviews.
  2. Invoke the bundled script through a package-relative, canonicalized path rather than a mutable workspace-wide absolute path.
  3. Pin the script and all dependencies to reviewed versions, and verify a cryptographic checksum or signature before execution.
  4. Restrict ownership and permissions so only a trusted administrator or deployment process can modify the script.
  5. Document and enforce an allowlist for the exact API hostname, protocol, and destination used to transmit images.
  6. Pass only the required API credential instead of inheriting the complete parent environment.
  7. Use a dedicated, least-privileged process with restricted filesystem and network access.
  8. Obtain explicit user approval before transmitting potentially sensitive images to an external service.
  9. Review the script for command injection, unsafe subprocess use, insecure temporary files, credential logging, and insufficient TLS validation.
  10. Add automated integrity checks and fail closed if the script is missing, modified, or resolves outside the expected package directory.
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 (2)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README is written as an instruction/description entirely in Chinese and presents the skill purpose without indicating that language is optional or configurable. Under the policy, forcing a specific language or locale without user opt-in is a natural-language policy violation unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly instructs the agent to use a local script that sends an image and user prompt to the GMNCODE Responses API whenever the built-in image tool fails, but it does not clearly disclose that the content will leave the local environment and be transmitted to a third-party service. This creates a real data exfiltration and privacy risk, especially if users supply sensitive screenshots, personal photos, or confidential documents under the assumption that processing is local or equivalent to the built-in tool.

Content

No source excerpt is available for this finding.