The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to bypass security boundaries and operate in an unrestricted mode.
The instructions mandate running commands "unsandboxed" because the platform sandbox allegedly hides necessary system resources.
It specifically directs the agent to "request escalated permissions" when using certain tools (e.g., Codex).
The companion script is designed to pass the --dangerously-skip-permissions flag to the underlying CLI tool.
[COMMAND_EXECUTION]: The skill relies on shell command execution to perform its primary function.
It executes a local companion script using Node.js: node "<skill-dir>/../../companion/agy-companion.mjs".
The command execution is intended to edit the local working tree directly with "unrestricted" permissions.
[CREDENTIALS_UNSAFE]: The skill requires access to sensitive authentication materials stored on the host system.
The documentation states that the tool needs access to an "OAuth token file" to authenticate with Google services.
By requesting to run unsandboxed, it gains access to files typically protected from the AI agent's environment.

agy-implementer