The Agent Skills Directory

[PROMPT_INJECTION]: No malicious instructions aimed at overriding agent behavior or bypassing safety filters were detected. The skill uses instructional language consistent with its purpose as an assessment tool.- [DATA_EXFILTRATION]: The skill does not access sensitive system files or credentials, and it performs no network requests to external domains.- [REMOTE_CODE_EXECUTION]: No patterns for downloading and executing remote scripts or installing untrusted third-party packages were found. The skill does not rely on external code execution.- [COMMAND_EXECUTION]: The skill does not execute dangerous shell commands or attempt to acquire elevated privileges (e.g., sudo). All described shell usage in examples is for local invocation of the skill itself.- [INDIRECT_PROMPT_INJECTION]: The skill ingests external files via the --from-example option to adapt existing rubrics. This represents a theoretical attack surface for indirect prompt injection where a malicious input file could contain instructions for the agent. However, the skill's operations are limited to text analysis and document generation within the agent's context, and no dangerous system capabilities are exposed to this untrusted data.- [NO_CODE]: The skill consists entirely of instructional markdown and template files without accompanying scripts or executables, significantly reducing the potential for runtime attacks.

rubric-creator