The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of agents that ingest and process untrusted user input directly into the model context.
Ingestion points: The Python example uses user_input = input("You: ") to gather data from the terminal which is then passed to Runner.run.
Boundary markers: While the skill imports RECOMMENDED_PROMPT_PREFIX, it does not implement explicit delimiters or instruction-guarding techniques for the interpolated user content in the provided code snippets.
Capability inventory: The skill defines tools such as lookup_tool and update_record. While currently limited, the multi-agent framework allows for complex tool-calling capabilities that could be influenced by malicious user input.
Sanitization: No input sanitization or validation logic is present in the provided templates.
[EXTERNAL_DOWNLOADS]: The Go implementation documentation references an external, third-party GitHub repository for agent functionality.
Evidence: Reference to github.com/nlpodyssey/openai-agents-go in the Go Agent Flow section. Users should verify the security posture of this community implementation before inclusion in production systems.

openai-agents