The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the Bash tool by interpolating prompts into a command string (e.g., gemini -p "<english-prompt>"). This pattern is vulnerable to command injection if the prompt contains shell metacharacters like double quotes or semicolons that are not properly escaped.
[DYNAMIC_EXECUTION]: The skill utilizes the gemini CLI tool which supports high-risk modes such as --approval-mode auto_edit and --yolo. While the instructions state these should only be used with explicit user consent, they represent a mechanism for the external tool to dynamically modify the local environment or execute actions based on AI output.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the analysis of external, potentially untrusted data such as code repositories and documentation by passing them to an external Gemini model.
Ingestion points: Data from the current project or codebase is read and included in prompts sent to the CLI tool (Step 2 in SKILL.md).
Boundary markers: The skill includes instructions to treat results as untrusted and to seek user confirmation before execution, providing a manual check but no technical isolation or delimiters for ingested data.
Capability inventory: The skill uses Bash, Read, and Write tools.
Sanitization: There are no documented steps for sanitizing external data or escaping shell metacharacters before it is interpolated into the CLI prompt.

gemini