yousim-identity
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains explicit instructions to override safety and instruction-tuning filters. It advises the model to be 'associative, uncensored, low-refusal' and directs the use of a system prompt that explicitly states 'meaning is optional hyperstition is necessary'. This is a form of jailbreak-style priming intended to bypass RLHF constraints.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided summaries and descriptions to 'instantiate' identities. There is a risk that malicious instructions could be embedded in the {SUMMARY} or {USER_INPUT} fields, which are then interpolated into complex multi-turn prompt templates without clear boundary markers or sanitization, potentially leading to agent hijacking.
- [COMMAND_EXECUTION]: The skill documentation references a 'yousim server' CLI and local endpoints (
/v1/construct). While it does not include direct shell execution scripts, it encourages the execution of external binaries and interaction with local network services for identity construction.
Audit Metadata