The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface via scripts/render_rubric.py, which ingests untrusted JSON data from the user. However, the script implements strict schema validation (requiring specific fields like domain, task, and dimensions) and is limited to string processing and markdown rendering. It does not possess capabilities for network access, arbitrary code execution, or sensitive file system operations, mitigating the risk of indirect injection exploitation.
[COMMAND_EXECUTION]: The documentation includes example bash commands for running the local scripts/render_rubric.py. These commands are standard for tool usage and do not involve obfuscated payloads, shell piping from remote sources, or privilege escalation (e.g., sudo).
[SAFE]: The skill's own code and instructions do not contain hardcoded credentials, obfuscated content, or persistence mechanisms. All logic is contained within standard Python and Markdown files, and the repository's internal trainer artifacts confirm that the skill underwent automated testing and optimization within a restricted environment.

judge-rubric