safety-synthetic-witness-flagger

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions and metadata contain no evidence of malicious intent, obfuscation, or dangerous operations. The logic is entirely focused on providing a forensic checklist for evaluating witness statements.
  • [PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted text documents (depositions, affidavits), which is a common vector for indirect prompt injection.
  • Ingestion points: Witness statements and deposition transcripts provided to the agent for triage (referenced in SKILL.md).
  • Boundary markers: There are no explicit instructions to the agent to disregard instructions that may be embedded within the witness testimony.
  • Capability inventory: The skill has no access to sensitive capabilities; it does not utilize file system tools, network requests, or subprocess execution.
  • Sanitization: The skill does not implement text sanitization, but given the lack of executable capabilities, the potential impact of an injection is negligible.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 04:28 AM
Security Audit — agent-trust-hub — safety-synthetic-witness-flagger