Follow the operation
- Keep the operation ID returned by a work request. An accepted or queued result means work is recorded, not complete.
- Read the operation through getOperation after a reconnect. Use If-None-Match with its ETag to avoid downloading an unchanged record. Prefer: wait may wait for a change within the server's returned limit.
- Read the final state, stages, effects and partial flag. Report succeeded, failed or cancelled exactly as returned. Cancellation cannot undo effects that already happened.
A request timed out
Send each new work request with a new Idempotency-Key. If its response is lost, repeat the identical request with the same key. Changing the key can start new work.
Keep the method, path, body and If-Match value unchanged on a retry. If the key was reused with different input, correct the client instead of retrying blindly. After the key has expired, inspect existing work before sending another request.
The outcome is unknown
outcome_unknown means RustSC could not prove what happened. It does not mean the action failed. Read the uncertainty and inspect the target. Do not send another restart, wipe, restore or console command just to get a clearer answer.
The current API reference lists the recovery operations actually available. Do not invent a resolve or retry endpoint from a planned workflow. If the current interface offers no suitable recovery, keep the operation ID and seek support.
Choose the next step from the error
- unauthenticated: sign in again. csrf_token_invalid: reload the session and repeat with its new token.
- forbidden or insufficient_scope: obtain the needed authority from a workspace owner or the integration consent flow when it exists. Never widen scope yourself.
- not_found: check your selected workspace and use an ID from a resource you can read.
- revision_conflict: read the current resource and compare it with the intended change before submitting again.
- operation_conflict: follow the blocking operation or use the action's documented queue option.
- capability_unavailable: read the reason and the server's available actions. Repeating a request cannot add a missing host or plugin capability.
- rate_limited: wait for Retry-After. dependency_unavailable: follow the error's suggested wait. Retry a work request with the same key.
- validation_failed: correct the listed field pointers. Keep the traceId for a support request; do not share credentials or full private responses.
Lists, limits and stale information
Use the documented cursor and filters for each list. A page of results is not the whole workspace. Do not assume every endpoint uses the same pagination fields.
Respect rate-limit responses. Check when a status was last observed and ask for a refresh if you need a current reading.