silent-failure-hunter

Pass

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to ingest and process untrusted external data (source code).
  • Ingestion points: The skill utilizes Read, Glob, and Grep to access and analyze the contents of external files.
  • Boundary markers: The instructions lack requirements for delimiters or specific system instructions to ignore commands potentially embedded within the audited code.
  • Capability inventory: The skill is granted access to high-privilege tools including Bash, TodoWrite, and Task.
  • Sanitization: There is no evidence of logic intended to sanitize or escape file content before it is processed by the AI agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 28, 2026, 11:35 PM