ralph

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • PROMPT_INJECTION (LOW): The skill presents an indirect prompt injection vulnerability (Category 8) by processing untrusted data to perform file operations.
  • Ingestion points: Input PRD markdown or text provided by the user.
  • Boundary markers: No delimiters or warnings are used to separate untrusted PRD content from agent instructions.
  • Capability inventory: Reading and writing prd.json, reading progress.txt, creating directories, and copying files within the local workspace.
  • Sanitization: The skill does not describe any methods for validating or escaping content from the ingested PRDs.
  • File System Access (SAFE): The file operations (read, write, archive) are scoped to the project's local directories and directly support the skill's stated purpose of project tracking and format conversion.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:46 PM