The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions provide a structured workflow for task decomposition and artifact generation. There are no attempts to override agent safety guidelines, bypass instructions, or extract system prompts.
[DATA_EXFILTRATION]: No network operations or commands (like curl or wget) are present. The skill operates locally on project files and does not reference external domains.
[COMMAND_EXECUTION]: The skill specifically instructs the agent to stop before code changes or implementation begin. No shell commands or subprocess execution patterns were identified.
[REMOTE_CODE_EXECUTION]: There are no external dependencies, package installations, or remote script downloads in any of the analyzed files.
[PROMPT_INJECTION]: The skill processes external PRDs, which represents an indirect injection surface. However, because the skill lacks network or execution capabilities, the risk of exploitation is negligible. Boundary markers and sanitization are not explicitly defined, but the limited scope of action (file writing only) maintains a safe posture.

ralph-plan