ralph
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill follows a workflow where user-supplied requirements are used to generate functional acceptance tests which subsequently direct the agent's code implementation tasks.
- Ingestion points: User requirements gathered in Phase 1 and stored in .agent/tests.json.
- Boundary markers: Not present in the instructions.
- Capability inventory: File writing, code implementation, and manual verification via tools.
- Sanitization: No sanitization of the requirement strings is described before they are processed by the agent.
- [PERSISTENCE]: The skill documentation mentions an afterStop hook called ralph-continue.sh. This mechanism is intended to re-trigger the agent if tasks are incomplete, effectively creating a persistent loop to ensure the feature build is finished, which is appropriate for the skill's iterative design.
Audit Metadata