Harness Builder (Interactive)

Goal

Guide the user through designing a small, single-purpose harness for a recurring task on their own project — a debt review, a security check, a deploy checklist, a bug-fixing routine, or anything else that follows a repeatable process.

A harness is not a single skill. It's a small workflow, built from whichever combination of pieces the task actually needs: a skill (something the agent runs on request), a rule (context the agent should always have, added to CLAUDE.md), a hook (something that must execute automatically at a specific event, not just be read), or occasionally a dedicated sub-agent. Most small harnesses need only one or two of these — do not reach for all of them by default.

This skill is conversational, not one-shot. Never assemble the final harness from a single guess at what the user wants. Walk through each decision below one at a time, and write nothing until the user has actually answered it.