docs: clarify fresh-session isolation from same-account Copilot Memory

[dliu120]

Question

Could you clarify the supported isolation contract for independent tasks run under the same authenticated GitHub account, using first-party Copilot inference rather than BYOK?

For each task, assume a new CLI process with a fresh private HOME / COPILOT_HOME, an empty per-session filesystem provider, and explicit session.create (not resume or fork). The relevant session settings are:

{
  "memory": { "enabled": false },
  "enableSessionStore": false,
  "skipEmbeddingRetrieval": true,
  "embeddingCacheStorage": "in-memory",
  "enableConfigDiscovery": false,
  "enableSkills": false,
  "enableFileHooks": false,
  "hooks": false,
  "tools": [],
  "availableTools": [],
  "mcpServers": {}
}

Remote export and remote control are disabled, and the request supplies no prior-conversation handles or imported history.

1. Do these controls prevent retrieval and use of pre-existing repository facts, user preferences, and earlier conversations—including any service-side context selection keyed by the same account or session metadata?
2. Does the session-level memory.enabled: false control cover the repository/user memories described by Copilot Memory, or is an additional account-level setting required?
3. If the account-level Copilot Memory setting must be Disabled, does that prevent reading/using existing memories, rather than only preventing new memories from being saved?

The distinction I am trying to understand is fresh local session state versus any additional context the first-party service may select before inference. A pointer to the applicable documentation, supported configuration, or version limitations would be helpful.

This is a documentation/contract question, not a report of observed cross-session leakage. I am not asking about deleting stored data or zero-retention guarantees.