sadd:do-in-parallel

To run parallel tasks across multiple files, dispatch independent sub-agents simultaneously to operate on each target, then validate outputs using meta-judge verification and automatic retries to ensure quality.

Meta-judge verification is an LLM-as-a-judge evaluation step applied after each parallel target completes. It enforces consistent evaluation criteria, provides structured judge feedback, and triggers automatic retries when targets fail quality checks.

You can scale batch work by dispatching parallel sub-agents to execute identical tasks across different targets simultaneously. This approach enforces per-target isolation and deterministic task segmentation to dramatically reduce total turnaround time.

Yes, parallel sub-agent dispatch supports automatic retries. It handles failed targets with sequential retry processing and per-target isolation, ensuring that retrying one target does not block or interfere with other parallel operations.

The best way to evaluate identical tasks across different targets is using a shared meta-judge evaluation plan with per-target judging. This ensures consistent evaluation criteria and structured feedback across all parallel sub-agent outputs.

You should avoid parallel sub-agents for task automation when targets have dependencies on each other, because this approach enforces per-target isolation and deterministic task segmentation designed specifically for independent implementations.