The Agent Skills Directory

[PROMPT_INJECTION]: The instructions explicitly direct the system to "always spawn with mode: 'bypassPermissions'" when performing analysis, which is an attempt to override platform safety constraints.
[CREDENTIALS_UNSAFE]: By instructing the agent to use Glob and Grep across the entire project structure while permissions are bypassed, the skill creates a high risk of exposing sensitive files such as .env files, SSH keys, or cloud credentials stored in the codebase.
[COMMAND_EXECUTION]: The skill requests execution with a bypass mode, aiming to circumvent established security boundaries and permission models of the agent environment during file system operations and codebase analysis.

idea