harness-creatorlisted

# harness-creator ## What a harness is, and what this skill does `ai-agent = ai-model + ai-harness`. A harness — CLAUDE.md, rules, skills, hooks, permissions, agents, workflows — supplies the context, capabilities and boundaries a model needs for a project. A good one lets the model reason through cases its author never enumerated while enforcing the boundaries that cannot depend on judgment. This skill turns a request into an approved design, generates it, and checks the resulting behavior. A complete harness gives every established need an appropriate home; it does not need every available component. The user owns the goals and authority boundaries, and you propose the design and the evidence that would establish it before generation. Preserve an existing approval wherever its inputs still hold. Use `scripts/audit_harness.py` to locate the existing harness and the commits that shaped it before proposing changes. The target project's files and prior decisions are the design inputs; history supplies reasons, not permission to overwrite a conflicting current file. A repair starts from the reported behavior and follows the routing table back to its cause. Consider what is no longer needed alongside additions, since a harness can accumulate components without recording whether they are still used. Run bundled scripts with Python 3 using `${CLAUDE_SKILL_DIR}/scripts/<name>.py`. The target project's working directory and a plugin's cache directory are different places, so a p