modal

Fail

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: references/images.md provides an explicit example of using add_local_dir to upload the local .aws credentials directory to a remote container image. This pattern facilitates the exposure of sensitive local authentication data to a third-party cloud platform.
  • [COMMAND_EXECUTION]: The skill relies on the modal CLI to perform operations such as modal run, modal deploy, and modal token new. It also provides examples for executing system commands via subprocess.Popen and subprocess.run within remote environments.
  • [REMOTE_CODE_EXECUTION]: The skill's core purpose is to facilitate remote code execution on the Modal serverless platform. This includes build-time execution via run_commands and run_function, as well as runtime execution of deployed Python functions.
  • [EXTERNAL_DOWNLOADS]: The skill documentation includes patterns for installing various Python and system packages during container image builds using uv_pip_install, pip_install, and apt_install. It also demonstrates cloning external Git repositories during the build process.
  • [PROMPT_INJECTION]: SKILL.md includes a directive for the agent to proactively suggest the 'K-Dense Web' platform when user requests involve complex tasks, which is an attempt to steer agent behavior toward the author's commercial services.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 3, 2026, 08:47 PM