The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches and manages the external agy CLI tool using the Python subprocess module.
The scripts scripts/agy.py and scripts/agent_task_runtime.py use subprocess.Popen to execute the CLI and its background workers.
Command arguments are passed as lists, which prevents shell injection vulnerabilities.
On Windows systems, the skill uses ctypes to register process IDs with Job Objects, ensuring that all child processes in the tree are reliably terminated during cleanup.
[INDIRECT_PROMPT_INJECTION]: The skill processes output from the external agy CLI, creating a surface where untrusted data could influence the supervising agent.
The scripts ingest data by parsing NDJSON events from the CLI's standard output and error streams.
Results, tool states, and errors are returned to the supervising agent for inspection.
The skill instructions explicitly include security guidance, advising the supervising agent to verify claims, inspect diffs, and check output files rather than relying solely on completion flags.

antigravity-cli