The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/humane_setup.py script uses subprocess.run(shell=True) to execute installation commands. These commands are constructed by embedding configuration values (such as paths for token_base) read from the project's humane.json or the user's global config. Because these values are not escaped or sanitized, an attacker can include shell control characters to execute arbitrary commands. The script relies on user confirmation ([y/N]) before execution, but a user following the tool's own diagnostic advice may be misled into approving a malicious command.
[REMOTE_CODE_EXECUTION]: The script uses dynamic code loading by modifying sys.path to import the illustrate module from a relative directory. Furthermore, the vulnerability to command injection allows for the execution of code provided by an untrusted project repository.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the agent-browser tool from the NPM registry and the addition of other skills from GitHub using npx. These actions download and execute external code on the host machine.

setup