The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze large external datasets such as entire codebases and architectural plans. When combined with automated approval modes like --approval-mode yolo, there is a potential for malicious instructions embedded in the analyzed code to influence the agent's behavior.
Ingestion points: Data is ingested from files, directories, and project documentation via the gemini command and the --include-directories flag.
Boundary markers: The provided command templates do not include explicit delimiters or instructions for the model to ignore embedded commands within the analyzed data.
Capability inventory: The skill utilizes the gemini CLI tool which has the capability to perform complex reasoning and code edits (via auto_edit and yolo modes). It also uses shell utilities like ps, pkill, kill, and timeout for process control.
Sanitization: There is no evidence of input sanitization or filtering of the content being processed by the CLI tool.
[COMMAND_EXECUTION]: The skill uses local shell commands to operate the Gemini CLI and manage its execution lifecycle. This includes process monitoring (ps), process termination (pkill, kill), and execution constraints (timeout).

gemini