Coordinated workers
A lead run divides a large outcome into bounded work packages, assigns them to isolated workers, and returns one reviewed result.
A single run can complete a task. Opulent can also operate the portfolio around it. The lead preserves the shared evidence contract, monitors progress, resolves overlap, and returns one reviewed result rather than a folder of unrelated drafts.
The lead can give each child a specific prompt, Skill, tag, model, and compute limit; inspect status and recent work; send a correction or clarification; stop work that is stuck or no longer needed; wait for a group of sessions to settle; and resolve overlap before outputs are combined.
Good parallel work has clear boundaries
- Auditing independent services.
- Researching separate questions.
- Analyzing different datasets.
- Drafting independent sections.
- Migrating packages that do not touch the same files.
Overlapping work adds coordination cost. Split by independent output, name the shared evidence contract, and decide how the parent run will combine results before launching child work. When the split is unclear, ask Opulent to propose the work packages first, naming each child's scope, shared context, dependencies, output contract, validation method, and merge order.
This is governed coordination rather than ambient access across every run. Each worker receives the context and authority required for its package, while the lead owns the combined outcome and the review boundary.
The shared evidence contract
Give every child the same return format for artifacts, evidence, status, and exceptions. Without it, the lead spends its effort interpreting five different shapes instead of compiling one result, and the reviewer inherits that inconsistency.
Each worker returns: artifact: the file or system change it produced evidence: the checks it ran and their output status: complete | blocked | partial exceptions: anything it declined to decide, with why