The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process response messages from a specified 'InfrastructureExpert'. This represents a data ingestion point where external input influences the agent's subsequent actions.
Ingestion points: The response content returned by the request tool (SKILL.md).
Boundary markers: The current implementation does not define specific delimiters or instructions to treat the incoming message as untrusted data.
Capability inventory: The instructions state that returned values should be used to 'configure code or tests,' which involves potential file-system writes or shell command execution.
Sanitization: There are no explicit instructions for the agent to validate or sanitize the connection strings and endpoint URLs received from the external source.
Sensitive Data Handling: The primary purpose of this skill is the retrieval and application of infrastructure secrets, such as database connection strings. While this is the intended behavior, it places sensitive credentials within the agent's context, requiring careful management during the execution of tasks.
Access Control and Guardrails: The skill implements a security best practice by restricting the agent's ability to run provisioning commands directly (e.g., az commands). This ensures that resource modifications are handled through a central, mediated request process rather than direct execution by the agent.

team-request