ralph-methodology

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
  • [Indirect Prompt Injection] (MEDIUM): The 'Ralph Loop' methodology reads instruction prompts from a local file (.claude/ralph-loop.local.yml), which can be modified by external processes or other agent actions.
  • Ingestion points: .claude/ralph-loop.local.yml (YAML state file).
  • Boundary markers: Absent; instructions direct the agent to re-feed the prompt content directly into the loop.
  • Capability inventory: 'Autonomous AI coding loops' imply the ability to execute system commands and modify files.
  • Sanitization: None provided; the agent is instructed to trust the prompt content in the state file.
  • [Command Execution] (LOW): The skill includes a bash command example (grep) used to monitor the iteration status of the loop.
  • [No Code] (INFO): The skill provides logic and architectural guidance but does not contain actual scripts, tool definitions, or executable code.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 07:37 AM