The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/executor.py script executes arbitrary shell commands retrieved from a YAML installation plan. It uses subprocess.run(command, shell=True), which allows for the execution of any shell command defined in the installation_plan.yaml file, including those involved in installation, verification, and rollback.
[COMMAND_EXECUTION]: The scripts/doc_generator.py script constructs shell commands by directly interpolating the target parameter (e.g., {target} --version). Because this input is not sanitized, an attacker could provide a target name containing shell metacharacters (e.g., ;, &&) to achieve command injection when the script is run.
[COMMAND_EXECUTION]: The scripts/plan_generator.py script similarly interpolates target and dependency names into shell command templates (e.g., f"brew install {target}"). These commands are then written to the installation plan, which is later executed by the agent or user.
[REMOTE_CODE_EXECUTION]: The skill's primary instructions and documentation (SKILL.md and docs/README.md) explicitly promote the curl | bash installation pattern. This method is a known security risk as it bypasses local package manager verifications and executes remote scripts directly in the shell.
[DATA_EXFILTRATION]: The scripts/env_explorer.py script performs extensive system reconnaissance, collecting data on the operating system, shell environment, environment variables (PATH), user names, and installed tools. This information is saved to env_info.json, providing a detailed footprint of the local system.

install-script-generator