task-classification

Multi-agent routing starts with task classification, which estimates scope, risk, and verification needs to assign the correct agent sequence and gates for safe execution. It categorizes work into Trivial, Exploratory, Implementation, Complex, or Research levels.

High-risk changes like database migrations, authentication, or security updates require mandatory verification gates including tests, lint, integration checks, and security review before completion. Task classification enforces these gates to prevent unsafe multi-agent execution.

Task classification evaluates parallel team eligibility by checking if subtasks are independent with exclusive file ownership. When criteria are met, it recommends parallel execution, deferring final criteria decisions to team guidance for optimal multi-agent orchestration.

Reclassification is supported through a protocol that upgrades task classification when scope or dependencies expand during exploration or implementation. This ensures the agent sequence and verification gate requirements adjust dynamically to handle increased complexity.

Task classification is needed when determining whether a request is exploratory, implementation, or complex, and specifying required verification gates. It applies to read-only analysis, bounded development changes, and high-risk work involving billing, API contracts, or authorization.