orchestrating-subagents

What problem does it solve?

This Skill eliminates common inefficiencies and risks in AI coding workflows that use parallel subagents: wasted token costs from overlapping parallel work, weakened validation gates from underpowered reviewer or critic agents, and cluttered main context filled with raw subagent outputs that are hard to synthesize.

Core Features & Use Cases

• Tiered role mapping: Assigns the right model effort and agent type to each role (lightweight explorers for recall-bound tasks, high-effort fresh-context agents for precision-bound reviewer and critic roles) to balance cost and validation strength.
• Fan-out discipline rules: Prevents overlapping parallel work, sets optimal agent counts per wave, and scales targeted follow-up waves instead of speculative large initial deployments.
• Standardized subagent prompt contracts: Defines clear scope, deliverable structure, evidence requirements, and status labels for all delegation prompts to ensure consistent, usable returns.
• Context hygiene and fallback rules: Keeps the main workflow thread clean, and provides clear fallback procedures when subagent tools are unavailable to avoid silently passing off self-review as independent review.
• Use Case: When running a swarm-mode QA sweep or issue tracer, use this Skill to delegate low-cost explorer agents to map codebase areas, then route findings to independent high-effort reviewer and critic agents to catch errors without exceeding your token budget.

Quick Start

Use the orchestrating-subagents skill to delegate parallel lightweight explorer agents to map the codebase for the current reported issue, then send the collected findings to independent high-effort reviewer and critic agents for validation before implementing any fixes.