multi-agent-workflow

Parallel development lets you run multiple AI agents simultaneously in isolated git worktrees, each implementing the same task using different approaches—such as different algorithms or frameworks. An observer monitors progress, runs quality checks, and compiles a comparison report so you can objectively select the best solution.

Yes. Git worktrees create isolated branches for each worker agent, allowing concurrent development without conflicts. A manager branch synthesizes results, and automated quality checks (tests, linting) run on each implementation to enable fair comparison.

Multi-agent workflows let you implement competing approaches concurrently and evaluate them on objective metrics: test coverage, performance, code quality, and maintainability. The observer agent compiles quantitative and qualitative assessments so you choose based on data rather than intuition.

An observer agent tracks all worker progress, runs automated quality checks across implementations, and generates a comprehensive comparison report. The manager agent then evaluates the results and merges the winning solution into the main branch.

The Skill handles worktree setup, manager branch creation, instruction templates, and status tracking automatically. You define the task and assign different approaches to workers; the system manages git coordination and quality checks.

Feature development, algorithm exploration, and bug-fix experiments are ideal use cases. Examples include exploring validation approaches (Zod schema-based, functional, decorator-based) or comparing different architectural patterns for the same requirement.