The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes defensive instructions to mitigate both direct and indirect prompt injection. Hard Rule 4 requires the agent to treat all codebase contents as inert data, ignore any embedded commands like 'ignore previous instructions,' and instead report such attempts as audit findings. The static analysis flag for prompt injection is a false positive arising from this explicit security guidance.- [COMMAND_EXECUTION]: The skill utilizes read-only search tools (e.g., grep) and version control commands (e.g., git rev-parse) for codebase reconnaissance. It specifically forbids any mutating operations, software installations, or build processes, maintaining a strict advisory-only posture.- [DATA_EXFILTRATION]: The skill is designed to audit local source code without performing network requests. It does not target sensitive system locations or credentials, and its file-writing capabilities are limited to producing implementation plans in a designated local directory.

improve-animations