What problem does it solve? Running multiple independent coding tasks sequentially wastes time, while dispatching subagents without a plan causes file conflicts, wrong model choices, and wasted budget. This Skill provides a structured protocol for decomposing work, routing each task to the right model tier, and coordinating parallel subagents safely. ## Core Features & Use Cases - Effort Routing: Assigns each task to a model tier (haiku-batch, sonnet-low/medium/high, opus-high, smart-search) based on file count, complexity, and architectural impact. - Batching and Agent Caps: Groups same-shape tasks into single dispatches and limits concurrent agents (default 4) to control cost and conflicts. - Dispatch Ledger and Fix Loop: Logs every dispatch per campaign in .sixth-sense/dispatch-log/<slug>.md, resumes interrupted campaigns, and runs a capped 5-round escalation loop for failed tasks. - Use Case: When implementing a feature spanning 8 independent files, present a dispatch plan table, get user confirmation, then run 4 parallel subagents at appropriate tiers while logging progress for resumability. ## Quick Start Break this feature into independent tasks and dispatch parallel subagents with the right model tier for each one.
