What problem does it solve? Delegating work to subagents without clear rules wastes tokens on overlapping explorers, weakens validation gates, and pollutes the main context with raw transcripts. This Skill defines how to assign agent types, models, and effort levels per role so delegation stays cheap where it can be and rigorous where it must be. ## Core Features & Use Cases - Role-to-tier mapping: Assigns explorers to cheaper tiers while mandating the strongest model and high effort for reviewers and critics. - Fan-out discipline: Launches 2-4 parallel agents only for disjoint scopes, in a single message, scaling waves rather than width. - Subagent prompt contract: Requires every delegation prompt to state scope, deliverable structure, evidence bar with file:line references, status labels, and output bounds. - Use Case: When running a swarm-mode PR review, launch parallel explorers over disjoint directories, then hand the consolidated findings to a fresh-context reviewer that independently re-derives each claim before approval. ## Quick Start Ask the agent to orchestrate a swarm-mode task by launching parallel explorer subagents with scoped prompts and a fresh independent reviewer for validation.
