agentsop-http-tool-wrapping
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a protocol for agents to ingest and process data from external HTTP APIs (REST, GraphQL, RPC), which creates a potential surface for indirect prompt injection.
- Ingestion points: The skill facilitates fetching data from third-party or internal APIs (SKILL.md, OP-1, OP-6, OP-7).
- Boundary markers: The SOP recommends shaping responses to include only relevant fields and using structured JSON outputs to mitigate model confusion, though it does not explicitly mandate "ignore instructions" delimiters for all raw data.
- Capability inventory: Tools constructed using this SOP typically possess network request capabilities (via httpx) and potentially side effects like create/update/delete operations (SKILL.md, OP-9).
- Sanitization: The skill emphasizes response shaping (OP-7) and error translation (OP-8) to ensure the agent receives structured, actionable feedback rather than raw upstream data.
- [EXTERNAL_DOWNLOADS]: The skill references several external documentation sources and technical blogs to support its design patterns, including official documentation from OpenAI, Anthropic, LangChain, and Stripe. These references are used for educational purposes and target well-known domains. The skill also recommends the use of standard Python packages like Pydantic and httpx for implementation.
Audit Metadata