What problem does it solve? In a spec-driven development pipeline, tasks in tasks.md must be tagged with type, specialization, and behavior metadata before they can be dispatched to specialized parallel agents — but doing this by hand is error-prone and unverifiable. This Skill automates that classification step with a deterministic validation gate so only structurally valid categorizations ever land on disk. ## Core Features & Use Cases - Subagent-based task classification: Dispatches one Sonnet categorizer session over tasks.md and plan.md to tag every task with type, specialization, preserves_behavior, runtime_consumed, and tags per taxonomy v1. - Zero-AI validation gate: Runs validate-categorization.py to enforce coverage, closed-enum membership, and the general-cap max(1, floor(0.2n)); categorization.md is kept only on exit code 0 and reverted or removed on failure. - Governed pipeline integration: Appends one categorizer trace record to traces.jsonl, honors the after_categorize commit hook, and supports a workforce_depth: light mode that routes all tasks to a cap-exempt light-fallback sentinel. - Use Case: After running /speckit-analyze on a feature, invoke /speckit-categorize to produce a validated categorization.md that the downstream /speckit-agent-assign phase consumes to build the task/agent roster. ## Quick Start Run /speckit-categorize in a spec-kit repository after tasks.md and plan.md exist to generate a validated categorization of all feature tasks.
