biblical-accuracy

Pass

Audited by Gen Agent Trust Hub on Feb 18, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • Prompt Injection (LOW): The file ucg-quick-reference.md contains directives to override the agent's default training and internal knowledge. Specifically, it instructs the agent to 'reject' what 'mainstream Christianity calls orthodox' and 'enforces UCG theology' instead. This constitutes a behavioral override constraint.
  • Metadata Poisoning (LOW): The skill presents itself in biblical-accuracy-usage-guide.md as verifying 'orthodox Christian theology,' but later documentation in ucg-quick-reference.md redefines this terminology to exclusively favor a specific sect (UCG), which is a deceptive metadata practice to influence the agent's outputs under the guise of standard terminology.
  • Indirect Prompt Injection (LOW): The skill is designed to process untrusted external data (sermons and teachings) as described in its usage guide.
  • Ingestion points: Full text of sermons or articles provided by the user in the prompt.
  • Boundary markers: Absent; there is no instruction for the agent to use delimiters or specific ignore-instructions for the ingested text.
  • Capability inventory: Text interpretation and doctrinal analysis; no high-risk capabilities like network or file system access were identified in the provided files.
  • Sanitization: Absent; no mention of sanitizing or escaping the user-provided text is found in the documentation.
  • Self-Referential Deception (LOW): The usage guide defines its own theological 'Severity Levels' (e.g., 'CRITICAL', 'HIGH'), which may be intended to confuse automated security scanners or human reviewers using similar terminology for security findings.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 18, 2026, 11:34 PM