The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes robust defensive measures against indirect prompt injection attacks. Specifically, 'Hard Rule 4' explicitly instructs the agent to treat file contents as data rather than instructions and to ignore/flag any attempts within the codebase to steer behavior (e.g., strings like 'ignore previous instructions'). The static detection of these patterns is a false positive reflecting a security best practice.
[COMMAND_EXECUTION]: The workflow utilizes standard, read-only shell commands such as grep and git rev-parse for codebase reconnaissance. Security is reinforced by strict operational rules that prohibit modifying source code, performing builds, or executing mutating operations.
[SAFE]: The skill operates entirely within a local, read-only context for the analysis phase and restricts all file-writing activities to a dedicated plans/ directory. No network operations, credential usage, or suspicious external dependencies were identified.

improve-animations