icp-builder-by-rdd
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a Python script (
scripts/build_child.py) and instructs the user to run it via the command line to generate output. The script is self-contained, using only the Python standard library for file operations and string templating without any external network dependencies. - [DYNAMIC_EXECUTION]: The core functionality involves the programmatic creation of new agent skill packages (including
SKILL.mdandopenai.yaml). The generation logic uses string templates and a dedicated sanitization function (md_text) that escapes Markdown markers and HTML entities to prevent data from interfering with agent instructions. - [INDIRECT_PROMPT_INJECTION]: The skill processes business data and customer stories provided by the user to build instructions for the 'child' agent. While this creates an attack surface for indirect prompt injection, the risk is addressed through character escaping and a strict data classification schema that distinguishes between observed facts, reported claims, and hypotheses.
Audit Metadata