The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill provides functionality to import API definitions from external URLs (e.g., using ContentFormat.OPENAPI_LINK). If an agent processes an untrusted or compromised URL, the imported content could serve as a vector for indirect prompt injection, potentially influencing the agent's behavior.
Ingestion points: The api.begin_create_or_update method in SKILL.md is shown fetching definitions from a remote URL.
Boundary markers: No specific boundary markers or instructions to ignore embedded commands are demonstrated in the snippets for processing these external definitions.
Capability inventory: The skill possesses extensive management capabilities over Azure infrastructure, including API configuration, user management, and policy updates, as detailed in the Operation Groups sections of SKILL.md and references/non-hero-scenarios.md.
Sanitization: While the Azure service performs its own validation, the agent's context may be influenced by the data returned from the external link before or during processing.
[Sensitive Information Handling]: A code example in SKILL.md demonstrates printing a subscription's primary key to the console. While intended for demonstration, this practice can lead to the accidental exposure of secrets in logs or process output if adopted in production environments.
Evidence: The "Create Subscription" example in SKILL.md ends with print(f"Subscription key: {subscription.primary_key}").

azure-mgmt-apimanagement-py