The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gemini.py executes an external command named gemini using subprocess.Popen. By passing arguments as a list rather than a shell string, the skill effectively mitigates the risk of shell injection through the prompt or model parameters. The script also includes a feature to change the working directory using os.chdir, which is handled safely through standard Python library functions.
[PROMPT_INJECTION]: The skill is a bridge to an external AI model, which creates a surface for indirect prompt injection if the data being analyzed contains malicious instructions.
Ingestion points: Data enters the system via sys.argv[1] (prompt) and sys.argv[2] (working directory) in the scripts/gemini.py file.
Boundary markers: The script does not implement delimiters or specific 'ignore instructions' markers when passing the user prompt to the Gemini CLI.
Capability inventory: The skill has the capability to execute the gemini binary and navigate the local filesystem via directory changes.
Sanitization: No input sanitization or escaping is performed on the prompt or the directory path before they are processed by the script.

gemini