Home / Docs / Review context before switching engines

Workflow guide

Review context before switching engines

Resume compatible sessions natively; transfer reviewed evidence when ownership changes.

Pre-release. Reviewed October 23, 2026. This review covers code behavior, not a packaged release or provider certification.

Keep the source identity

Record the source provider, account and session, then the destination provider/account. Native resume requires compatible ownership. A cross-provider handoff creates a new destination session; it does not turn a foreign native session into a resumable one.

Inspect the bridge

  1. Finish the active turn and keep its conversation intact.
  2. Review the proposed transition's candidate provenance, exclusions, privacy decisions and bounded context.
  3. Remove unrelated evidence. A graph relationship is not permission to copy everything.
  4. Approve the exact current preview, or choose a clean start with no bridge payload.
Source

Owned session and current evidence revision.

Human review

Bounded preview, exclusions and privacy decision.

Destination

Explicit new identity and delivery receipt.

commandem transition plan --help
commandem transition preview --help
commandem transition review --help

Use the installed source-session, destination-account and request-file flags. Do not invent a session ID or copy credentials into a request.

Confirm the delivery, not the model's understanding

Inspect the transition receipt and destination identity. Feedback can say whether transferred context helped, distracted or became stale. That feedback is distinct from a related/unrelated session-pair label and from independent verification.

Expected result and recovery

Changed source, destination or privacy evidence invalidates an old review. Rebuild it; do not force the stale payload through. A clean start transfers no bridge context. Preserve the source even when destination delivery fails.

Matrix is deterministic when no learned scorer is active. A graph index, labeling queue or training pipeline is not proof of an active GNN. Unknown provider training/retention settings remain unknown.