business-simulation

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill establishes an autonomous operating environment where the agent is instructed to verify data against external web sources. This creates a surface for indirect prompt injection, as malicious data from those sources could influence the agent's autonomous decisions and actions within the simulation.\n
  • Ingestion points: Web search results for grounding simulation entities (SKILL.md, Section 2: 'verified against a real source. Search first').\n
  • Boundary markers: Not present. The agent is not instructed to treat external data as untrusted.\n
  • Capability inventory: The agent is granted delegated authority to 'decide and execute' routine work autonomously (SKILL.md, Section 4).\n
  • Sanitization: Not present. External content is used to build the simulation's 'record of truth' without filtering.\n- [COMMAND_EXECUTION]: The skill includes a Python script template for calculating simulation days and checking for day-number drift. The agent is expected to maintain and execute this script locally as part of its operational loop.\n- [SAFE]: The provided methodology and utility script do not contain hardcoded credentials, persistence mechanisms, or malicious obfuscation techniques. The external source referenced in the metadata identifies a legitimate repository path.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 02:13 PM
Security Audit — agent-trust-hub — business-simulation