What problem does it solve?
Running the same or related tasks across many files sequentially is slow and error-prone, and unverified agent output often ships with defects. This Skill dispatches independent tasks to parallel sub-agents with per-task model selection and LLM-as-a-judge verification so batch work finishes faster and meets a quality bar.
Core Features & Use Cases
- Parallel sub-agent orchestration: Dispatches implementation agents concurrently across files or named targets after validating task independence.
- Requirement grouping: Identifies repeatable, shared, and independent task groups to reduce the number of meta-judge and judge agents without losing evaluation accuracy.
- Per-task model selection: Assigns haiku, sonnet, or opus tiers per task based on scope, complexity, and risk, with escalation rules on retry.
- LLM-as-a-judge verification: Meta-judges generate evaluation specs and judges score each result, with up to 3 retries per target.
- Use Case: Apply the same error-handling refactor across a dozen source files at once, with each file's change independently verified against a generated rubric before being accepted.
Quick Start
Ask the agent to run do-in-parallel with a task description and a comma-separated --files or --targets list, optionally adding --model or --strict flags.
