The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the terminal tool to execute various agy commands for plugin management, binary updates, and running model prompts.
Evidence: Examples provided in SKILL.md show commands like agy plugin list and agy -p '...' being executed via the terminal.
Evidence: Mentions the use of the --dangerously-skip-permissions flag which bypasses internal CLI safety checks.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions to download and execute installation scripts from official Google domains.
Evidence: references/cli-docs.md contains curl and irm commands targeting https://antigravity.google/ for script execution.
[REMOTE_CODE_EXECUTION]: Installation instructions include piping remote scripts directly into shell interpreters.
Evidence: references/cli-docs.md describes executing installation scripts from antigravity.google using | bash and | iex.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from log files and project repositories.
Ingestion points: The skill reads log files at ~/.gemini/antigravity-cli/log/cli-*.log and processes arbitrary project files when using the agy -p command.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the documentation.
Capability inventory: The skill has access to the terminal tool for executing shell commands and read_file for accessing the file system.
Sanitization: There is no evidence of sanitization or validation of the content read from external files before it is processed by the agent.

antigravity-cli