The Agent Skills Directory

Network Permission Escalation: The skill includes instructions to use sandbox_permissions=require_escalated when encountering transport failures such as DNS errors or timeouts when reaching api.openai.com. This is a functional requirement for troubleshooting connectivity in restricted execution environments. The guidance follows best practices by instructing the agent to apply this escalation only to the specific failing request and avoiding its use for unrelated tasks.
References to Official Services: The instructions provide links to platform.openai.com for billing and usage limit management. These links direct users to official vendor-owned domains to resolve quota-related errors and are consistent with the skill's purpose.
Indirect Prompt Injection Surface: This skill is designed to analyze error messages from API requests. Because these messages are external data that could potentially be influenced by an attacker (e.g., through a manipulated upstream service), they represent a standard ingestion surface. The skill manages this by providing specific routing rules based on concrete HTTP status codes and error constants.
Dynamic Diagnostic Execution: The evaluation cases described in the skill suggest scenarios where the agent may generate and execute temporary scripts to test API connectivity. This is a common pattern for developer-oriented AI agents and is used here specifically to verify transport paths before diagnosing API-side issues.

openai-api-troubleshooting