_plan-evaluation

Plan evaluation identifies and quantifies quality by checking task specificity, dependency ordering, completeness, feasibility, and researchability. It applies deterministic checks against artifacts like idea documents, codebase structures, and task lists to prevent ambiguity before execution.

Checking task dependencies involves enforcing correct dependency sequencing, maximizing parallelism, and avoiding circular dependencies. The evaluation verifies proper ordering against task lists to prevent execution bottlenecks.

A task plan is feasible when task sizing, agent capabilities, wave size, and domain gotchas are assessed. Feasibility evaluation checks these factors to ensure tasks match agent skills and avoid domain-specific pitfalls.

You can verify plan completeness by checking full scope coverage, defined verification criteria, and scope containment. The evaluation uses tools like Read, Grep, Glob, and Bash to inspect codebase structures and task lists.

Assessing researchability involves checking for clear research questions, bounded scope, codebase pointers, and external research needs. This ensures research tasks have well-defined objectives and necessary context before delegation.

Task plans suffer from scope creep and ambiguous deliverables when they lack explicit deliverables, proper dependencies, and measurable criteria. Plan evaluation applies a structured, repeatable method to prevent these issues before execution.