python-executor

Warn

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill requests permission to execute the infsh (inference.sh) command-line tool. This allows the agent to interact with remote infrastructure and manage remote applications.
  • [REMOTE_CODE_EXECUTION]: The core functionality is to run arbitrary Python scripts provided as strings. The execution environment includes network access and a wide array of libraries (e.g., requests, Selenium, Playwright), which could be misused to perform unauthorized network operations or interact with external services if the input code is malicious.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection.
  • Ingestion points: The code field in the input schema accepts raw Python strings generated or retrieved by the agent from potentially untrusted sources.
  • Boundary markers: There are no markers or validation rules defined to ensure the agent distinguishes between its own instructions and data retrieved from the web when constructing the Python code.
  • Capability inventory: The execution environment allows for subprocess execution, network requests, and file manipulation within the remote container.
  • Sanitization: The skill does not provide any mechanisms to sanitize or validate the Python code before it is sent to the executor.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 7, 2026, 12:04 PM