source-command-implement-batch

Batch processing for code migration is parallelized by parsing CSV rows and launching up to 5 concurrent sub-agents. The orchestrator assigns non-overlapping tail rows to each agent, ensuring safe parallel execution while continuously backfilling as agents finish their work.

Tail-based batch migration processes the last rows of a CSV file by assigning immutable offsets to agents and removing processed lines upon completion. Removing processed lines from the source-of-truth CSV signals progress autonomously and prints a continuation command for the next batch.

Multiple agent workflows run concurrently by assigning immutable row offsets from a single CSV input to each sub-agent. This deterministic assignment ensures non-overlapping work allocation, while the orchestrator aggregates structured per-agent reports including implemented, already-done, skipped-vanilla, and failed statuses.

Concurrency limits for automated card implementation are capped at 5 parallel sub-agents. Each sub-agent runs a filtered test for its specific card, while the orchestrator executes the broader CI check only after all agents finish to maintain batching discipline.

When a sub-agent fails during parallel code migration, the orchestrator logs the failure as a structured per-card outcome and continues processing the remaining batch. The system aggregates all agent reports, allowing successful migrations to complete even when individual cards fail.