What problem does it solve?
When you have multiple independent failures (such as unrelated failing test files, separate broken subsystems, or decoupled audit findings), investigating and fixing them sequentially wastes time and consumes unnecessary context. Each independent issue can be worked on in parallel without shared state or dependencies, drastically reducing total resolution time.
Core Features & Use Cases
- Dual fan-out modes: Supports remediation fan-outs for code/config edits and read-only evaluation fan-outs for audits, with mode-specific safeguards for each.
- Pre-dispatch feasibility checks: Validates that all target tasks are live, in scope, and not already completed or being worked on by concurrent actors to avoid wasted agent effort.
- Collision and concurrency safeguards: Includes rules for shared repository safety, including handling shared worktrees, commit-time index races, and namespace allocation to prevent write conflicts.
- Structured agent prompts: Provides templates for focused, self-contained agent tasks with clear scope, constraints, and output requirements to ensure consistent, usable results.
- Use case: If 3+ unrelated test files are failing with different root causes, use this skill to dispatch one specialized agent per test file to fix them concurrently, rather than investigating each one sequentially.
Quick Start
Use the dispatching-parallel-agents skill to fix all unrelated failing test files in the current repository in parallel.
