adr-verbatim

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input and writes it directly to the file system, which serves as a potential vector for indirect prompt injection if other agents or tools process these records later.
  • Ingestion points: User-provided text for the ADR body requested in step 1 of SKILL.md.
  • Boundary markers: No delimiters or protective warnings are included in the generated output to isolate user-supplied content from the rest of the document.
  • Capability inventory: The skill utilizes file system read access to determine ADR numbering and write access to create new files in the docs/adr/ directory.
  • Sanitization: No sanitization, filtering, or escaping is performed on the user input; the skill instructions explicitly mandate that the text must be recorded verbatim with nothing added, removed, or fixed.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 12:31 AM
Security Audit — agent-trust-hub — adr-verbatim