ai-security-trainer-2026

Fail

Audited by Gen Agent Trust Hub on Aug 4, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download source code from https://github.com/dylanhwiireed3200/ai-security-trainer-2026.git, which is an unverified third-party repository not associated with the author's known infrastructure.
  • [REMOTE_CODE_EXECUTION]: The installation process requires running pip install -r requirements.txt and python app.py on the downloaded content, leading to the execution of code from an untrusted source.
  • [COMMAND_EXECUTION]: The skill documentation includes multiple shell commands for setup and execution, including git clone, pip install, flask run, and docker run.
  • [CREDENTIALS_UNSAFE]: The application code in app.py contains a hardcoded fallback value for the SECRET_KEY ('dev-secret-key').
  • [DATA_EXFILTRATION]: The skill's frontend stores authentication tokens and user IDs in localStorage and transmits them via API headers, while the backend stores user data in unencrypted local JSON files.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection via lesson and user profile data ingestion.
  • Ingestion points: External data is processed via the /api/lessons and /api/users/<user_id>/progress endpoints.
  • Boundary markers: There are no delimiters or explicit instructions to ignore embedded commands in the processed data.
  • Capability inventory: The skill possesses capabilities for command execution, file system writes, and network operations.
  • Sanitization: The code uses basic JSON parsing without explicit validation or sanitization of the data fields before they are used in the application logic.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 4, 2026, 10:33 PM
Security Audit — agent-trust-hub — ai-security-trainer-2026