Essential guidelines

Before and after a run

What to check before substantial work begins, and what to change once it finishes.

Before a run

  • Does the job have a clear start, end, and owner, and is the expected artifact named?
  • Are success criteria explicit and checkable, and are allowed and excluded changes clear?
  • Are the correct files, links, data, designs, and system records attached?
  • Is the source of truth named when inputs conflict, and are the necessary integrations connected?
  • Which tests, checks, reconciliations, citations, or visual proofs should pass?
  • Which failure blocks completion, and what evidence should the reviewer receive?
  • Which findings can Opulent fix and recheck on its own, and which decisions require a person?
  • Can one run hold the goal and context without losing focus, or should the work be split?

A job is well scoped when one run can state the plan, own the artifact, and prove completion without inventing missing requirements. The number of hours is less useful than that test.

After a run

  1. 01Review the artifactOpen the deliverable and test it against the original criteria.
  2. 02Review the evidenceConfirm that sources, checks, screenshots, recordings, or system receipts support the completion claim.
  3. 03Review the pathLook for repeated retries, dead ends, unnecessary model use, missing context, and environment failures.
  4. 04Classify correctionsDecide whether each change reflects an error, a preference, a policy change, missing context, or a one-time exception.
  5. 05Update the right layerChange Knowledge, the Skill, a validator, a tool rule, routing guidance, or the next request.
  6. 06Measure the next runCheck whether the update reduced rework or created a new failure.

If a session repeatedly reaches a compute or context limit, reduce the scope or divide the job. If the environment caused the failure, fix the workspace before changing the prompt. If the same clarification appears in several runs, capture it as reusable context.

Habits that produce better runs

  • Start with a job, not an ambition. Make the first run narrow enough that another person can tell whether it finished.
  • Define completion in checkable terms, and say when a missing source or approval should stop the run.
  • Review the plan before expensive work, so several agents do not work from the wrong interpretation.
  • Parallelize independent work and keep coupled work in one coordinated run.
  • Inspect the artifact and the evidence. The completion message points to the work and cannot replace review.
  • Refine the system, not only the prompt.
  • Ship only what you understand. Opulent executes the work; your team remains responsible for what ships.

Treat generated work like work from any other contributor. Inspect it, test it, challenge its assumptions, and understand it before it goes out under your name.