agent-self-evaluation
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a rubric-based evaluation system. The logic is entirely contained within markdown instructions and a Python script that uses standard library modules for string processing.
- [COMMAND_EXECUTION]: The script
scripts/evaluate.pyperforms heuristic analysis using regular expressions on text input provided via stdin or local files. It does not invoke the shell, spawn sub-processes, or execute dynamic code. - [DATA_EXFILTRATION]: There are no network-capable functions (e.g.,
requests,urllib,socket) used in the scripts. The skill does not access sensitive system paths or environment variables. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (agent output and user requests). However, the analysis is limited to non-executable regular expression matching for the purpose of scoring, which does not provide a mechanism for system-level exploitation.
Audit Metadata