dispatching-parallel-agents_skill

Overview

This skill dispatches independent agents to investigate and fix multiple unrelated problems in parallel. It is designed for situations with two or more independent tasks that have no shared state or sequential dependencies. The approach speeds remediation by isolating scope and running concurrent investigations.

How this skill works

Identify independent problem domains and create a focused task for each one (one test file, subsystem, or bug). Dispatch one agent per domain with a self-contained prompt, clear goal, and explicit constraints. When agents finish, review their summaries, check for conflicts, run the full test suite, and integrate changes.

When to use it

• Multiple failing test files with different root causes
• Several subsystems broken independently
• Each problem can be diagnosed without context from others
• You want to reduce wall-clock time by parallelizing investigations
• No shared state or file contention between tasks

Best practices

• Group failures into independent domains before dispatching agents
• Write focused, self-contained prompts with specific expected outputs
• Give each agent clear constraints (e.g., do not modify unrelated files)
• Prefer event-based fixes over arbitrary timeouts for timing issues
• Review each agent summary and run the full test suite before merging

Example use cases

• Three test files fail after a refactor; dispatch one agent per file
• Independent subsystems (approval flow, batch processing, abort logic) each need debugging
• Parallel agents investigate unrelated race conditions across the codebase
• Speed up triage when multiple unrelated CI jobs fail simultaneously
• Assign different engineers or bots to focused test fixes concurrently