ralph-loop
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and follow instructions from external, potentially untrusted files such as
TODO.mdand.claude/ralph-loop.local.md. - Ingestion points: The agent is instructed to read
TODO.mdand.claude/ralph-loop.local.mdto define its current objective, tasks, and verification commands. - Boundary markers: The skill does not provide specific delimiters or 'ignore embedded instructions' warnings for the content within these task files, allowing the content to influence agent behavior directly.
- Capability inventory: The agent is expected to possess full coding capabilities, including file modification and shell command execution (e.g., test and build commands), which are directed by the task files.
- Sanitization: Content from the ingested state and task files is processed as instructions without escaping, validation, or filtering.
Audit Metadata