What problem does it solve? Agents often reach for a Python script that loops over items calling an LLM API, reinventing authentication, rate-limit handling, retries, and cost plumbing. This Skill redirects that instinct toward spawning parallel subagents inside the current session, which is faster to set up and requires no API key. ## Core Features & Use Cases - Sharded batch processing: Split a corpus into batches and run one subagent per batch for labeling, classifying, extracting, or summarizing many items, then merge structured results. - Role-based ensembles: Spawn one subagent per role with differentiated briefs for evaluator ensembles, multi-rater review, red-team checks, or multi-rubric scoring. - Clear boundary guidance: Distinguishes when subagents win (interactive, hundreds-to-thousands of items, judgment-heavy work) from the one real exception (unattended very-large batch jobs like nightly CI re-tagging of tens of thousands of rows). - Use Case: You need to label 2,000 documents with categories. Instead of writing a for doc in docs: client.messages.create(...) loop, shard the documents across parallel labeling subagents and merge their JSON output. ## Quick Start Ask the agent to label, classify, or summarize your batch of documents by spawning parallel subagents instead of writing an API-calling script.
