ralph-loop
Pass
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements an iterative loop mechanism that creates a surface for indirect prompt injection. Malicious data ingested or generated in one iteration is stored in a state file and re-processed in the next, potentially influencing agent behavior across iterations.
- Ingestion points: User-provided task prompt (SKILL.md) and state data from the
.cursor/ralph/scratchpad.mdfile. - Boundary markers: The prompt is wrapped in markdown formatting within the state file but lacks specific delimiters to distinguish user input from control instructions.
- Capability inventory: The agent has file system write access to create and update the
.cursor/ralph/scratchpad.mdfile. - Sanitization: No input validation or sanitization of the task prompt is implemented before storage.
Audit Metadata