simmer-generator

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection via its feedback-driven design.
  • Ingestion points: Processed artifacts ('Current candidate') and external feedback ('ASI'
  • Artificial Super Intelligence direction) are ingested as natural language instructions (SKILL.md).
  • Boundary markers: The instructions do not define boundary markers or delimiters to isolate untrusted data from the agent's control logic.
  • Capability inventory: The skill has the authority to write to the file system, modify existing scripts, and execute shell commands (SKILL.md).
  • Sanitization: No sanitization or verification protocols are mentioned for the external feedback before it is used to drive workspace changes.
  • [COMMAND_EXECUTION]: The skill is instructed to execute shell commands provided in the 'VALIDATION_COMMAND' field of the setup brief to verify infrastructure changes and pipeline integrity (SKILL.md).
  • [REMOTE_CODE_EXECUTION]: The skill is designed to modify evaluator scripts and infrastructure configurations ('evaluator scripts may be modified', 'switching models'), which results in the execution of agent-generated code during the evaluation cycle (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 01:30 AM
Security Audit — agent-trust-hub — simmer-generator