The Agent Skills Directory

Prompt Injection (SAFE): The skill uses directive language ("MANDATORY", "STOP") to enforce grounding in official documentation through the gemini-cli-docs skill. This is a defensive measure to ensure accuracy and does not attempt to bypass safety filters or extract system prompts.
Data Exposure (SAFE): While the skill discusses sensitive configuration paths like ~/.gemini/settings.json and ~/.gemini/trustedFolders.json, it does so in an instructional context for user configuration. There are no patterns suggesting unauthorized access or hardcoded credentials.
Command Execution (SAFE): The skill lists tools like Read and Grep but does not contain scripts that execute arbitrary shell commands or escalate privileges. References to shell execution are in the context of describing how the Gemini CLI handles 'Trusted Folders' security boundaries.
Remote Code Execution (SAFE): No external downloads or remote script executions were detected. It relies on internal skill delegation.
Indirect Prompt Injection (LOW): The skill has a data ingestion surface via Read, Glob, and Grep tools which could potentially process untrusted content within configuration files. However, the skill lacks write or network capabilities, and its instructions explicitly mandate basing responses on official documentation, which serves as a grounding boundary.

gemini-config-management