The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides explicit instructions to bypass host environment sandboxing. Specifically, the troubleshooting documentation advises the agent to request escalated permissions or have the user grant full workspace access if the tool's access to network ports or sensitive token files is blocked by sandbox security features.
[CREDENTIALS_UNSAFE]: The skill documentation identifies and targets sensitive authentication files, specifically OAuth tokens located at ~/.gemini/. It instructs the agent to ensure the companion script runs outside of a sandbox to successfully read these secrets.
[PERSISTENCE]: The setup process modifies a global configuration file on the user's machine to apply allow/deny rules. These changes are not scoped to the current project and affect the security posture of the tool across all projects on the machine.
[INDIRECT_PROMPT_INJECTION]: The skill processes results and diagnostics from external background jobs. Evidence: 1) Ingestion point: node ... agy-companion.mjs wait <id> in SKILL.md returns data to context. 2) Boundary markers: Absent; instructions tell agent to assess task completion based on the data. 3) Capability: SKILL.md (Bash), setup.md (config write). 4) Sanitization: No sanitization of ingested job results is described.
[COMMAND_EXECUTION]: The skill relies on executing a local companion script (agy-companion.mjs) via Node.js to perform its primary functions, including job management and system configuration.

jobs