A Copilot cloud agent is assigned an issue that includes a one-off production error, a temporary customer identifier, and a log excerpt that applies only to this incident. The team wants the agent to use that context for the current fix without making the customer identifier influence later work in the repository. Which memory strategy should you choose?

Keeping the details in task context is the correct strategy because it ensures the information is available only for the current task. This prevents the one-off production error, temporary customer identifier, and incident-specific log excerpt from being stored permanently or influencing future, unrelated work in the repository. Adding them to repository custom instructions or storing them as repository-level facts would make them persistent, potentially leading to unintended side effects or irrelevant data being used in subsequent tasks. Querying them through an MCP server is not a standard or appropriate memory strategy for managing temporary, task-specific context within a Copilot agent's workflow.