pydantic-ai-tool-system
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents how to create agent tools that process external input and perform powerful operations, establishing an attack surface where malicious data could influence agent behavior.
- Ingestion points: Multiple tool examples in
SKILL.md, such assearch(query: str),query_users(name: str), andupdate_user(id: int, data: dict), ingest potentially untrusted data from the language model's output. - Boundary markers: The code templates do not demonstrate the use of boundary markers or specific instructions to the model to ignore malicious sub-prompts within tool arguments.
- Capability inventory: The documentation provides examples of tools that interact with databases (
ctx.deps.db), use API clients (ctx.deps.api_client), and perform network I/O (ctx.deps.client.get('/data')) as seen inSKILL.md. - Sanitization: The provided examples lack implementation of input sanitization, validation, or escaping mechanisms to protect against command or query injection within the tool logic.
Audit Metadata