migrating-openai-agents-sdk-to-pydantic-ai
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external repository content including instructions, dependency files, tests, and code, creating a surface for indirect prompt injection. If a target repository contains malicious instructions, they could attempt to subvert the migration logic. The skill proactively addresses this by providing high-quality mitigation guidance, such as enforcing authorization at the service layer, using execution sandboxes (e.g., ModalSandbox), and distinguishing trusted application context from untrusted model-generated inputs.
- Ingestion points: SKILL.md (Step 1: Read repository instructions, dependency files, tests, and runtime entrypoints).
- Boundary markers: The instructions specify that RunContextWrapper.context is trusted application state and distinguish it from model-generated fields.
- Capability inventory: The migration flow references the use of tools for file manipulation (FileSystem), command execution (Shell, Coder), and network operations.
- Sanitization: Recommends using InputGuardrail, OutputGuardrail, and ToolGuardrail to enforce security policy below the model layer.
Audit Metadata