qinghu-model-face-swap

Fail

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the Node.js binary and associated checksums from the official nodejs.org website to satisfy environment dependencies. It also installs the vendor-supplied @iqinghu/qhkit package from the npm registry.
  • [PERSISTENCE]: Instructions guide the agent to modify the user's shell profile (~/.bashrc) to ensure the Node.js binary remains in the system PATH across sessions, which is a persistent modification of the host environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user instructions and interpolates them directly into the JSON payload for the face-swapping tool.
  • Ingestion points: User-provided prompt and uploadedImages in SKILL.md examples.
  • Boundary markers: None; user input is treated as raw data within the tool's JSON command structure.
  • Capability inventory: The skill executes the qhkit CLI tool, which communicates with external APIs to process media files.
  • Sanitization: The skill does not perform escaping or validation of user-provided prompt content before tool invocation.
  • [COMMAND_EXECUTION]: Executes shell commands to install software (npm install), manage file systems (mkdir, tar), and run the qhkit command-line utility.
Recommendations
  • HIGH: Downloads and executes remote code from: https://nodejs.org/dist/v22.22.3/SHASUMS256.txt - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 21, 2026, 04:09 PM