Skills
skills/naimalarain13/hackathon-ii_the-evolution-of-todo/openai-agents-sdk/Gen Agent Trust Hub

openai-agents-sdk

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (LOW): The templates require the installation of the openai-agents package, which is a third-party framework rather than the official OpenAI client library. Evidence: Installation instructions found in templates/agent_gemini.py, templates/agent_mcp.py, and examples/todo-agent.md.
  • [PROMPT_INJECTION] (LOW): The skill implements an architecture vulnerable to indirect prompt injection via the Model Context Protocol (MCP) surface. 1. Ingestion points: templates/agent_mcp.py and examples/todo-agent.md ingest tool definitions and execution results from an external MCP_SERVER_URL via the MCPServerStreamableHttp class. 2. Boundary markers: Absent. Agent instructions do not include delimiters or warnings to ignore instructions embedded in server-provided tool data. 3. Capability inventory: The agents can invoke tools provided by the connected MCP server, which may include state-changing operations like delete_task as seen in todo-agent.md. 4. Sanitization: No sanitization or schema validation of the data returned from the MCP server is present in the templates.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:09 PM