Continuing long work
The workflow owns the job beyond a single response, and continuity comes from compact checkable state rather than an unlimited transcript.
Opulent checks whether required work is actually satisfied, yields safely when an execution budget is reached, and resumes under the same outcome contract. A long job does not end because a response ended.
Handoff and compaction
When the working context must be compacted, Opulent carries forward an agent-authored handoff: verified state, completed work, blockers, remaining steps, and the next action. Superseded exploration is left behind.
That is a deliberate design choice. Carrying an ever-growing transcript makes a long run slower and less accurate, because the useful state gets buried in the record of how it was reached. A compact handoff that can be checked is more reliable than a complete history that cannot.
Pausing for authentication
If a tool requires OAuth, the run pauses, presents the connection step, and resumes the exact run after you authenticate. You complete the sensitive handoff; Opulent continues with the same workspace, evidence, and pending action.
When a run keeps hitting limits
If a session repeatedly reaches a compute or context limit, the answer is usually scope rather than configuration. Reduce the scope or divide the job into independent packages.
If the environment caused the failure, fix the workspace before changing the prompt. A prompt cannot compensate for a missing dependency or an unreachable system.