equivalence-class-audit

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is designed for read-only inspection and does not contain any commands for network access, file modification, or credential handling.\n- [SAFE]: Includes explicit defensive instructions for the agent to treat inspected artifacts as data and avoid executing any code or instructions found within them.\n- [SAFE]: Protects against obfuscation and injection by requiring NFKC normalization, HTML entity decoding, and the removal of Unicode bidirectional and format control characters before data comparison.\n- [SAFE]: Explicitly forbids the use of homoglyphs (mixing Latin and Cyrillic characters) in candidate labels, preventing script-based spoofing attacks.\n- [INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted content from the user and external files, it mitigates injection risks through its read-only design and mandatory normalization steps.\n
  • Ingestion points: User-supplied 'Triggering finding' and files within the user-defined 'Locked audit scope' (e.g., config, tests, documentation).\n
  • Boundary markers: Uses a structured report format with specific header labels (e.g., '## Equivalence-Class Audit Report') to maintain context separation.\n
  • Capability inventory: The skill is limited to read-only inspection; it explicitly forbids execution of artifacts or implementation of fixes.\n
  • Sanitization: Implements extensive sanitization including removal of bidi controls, decoding of HTML entities, and NFKC normalization.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 09:41 AM