skill-optimizer

Incremental skill optimization involves proposing small, single changes guided by an eval rubric to measure success. This approach minimizes risk and includes a rollback plan to preserve existing functionality during updates.

An eval rubric for skill quality assurance is a set of 3–6 measurable criteria covering trigger quality, task fidelity, safety, and output conformance. It establishes explicit success metrics to evaluate target skill improvements.

Audit a Claude Code skill by inspecting its current package, invocation settings, and examples. This baseline audit identifies concrete weaknesses and scopes changes to small, testable updates.

Yes, you can tighten a skill's prompts to reduce over-triggering while preserving functionality. Incremental optimization proposes single changes linked to an eval hypothesis to measure trigger quality and task fidelity.

Plan a rollback for a skill optimization update by defining risk management rules and a clear evaluation plan beforehand. This prevents regressions by ensuring failed incremental changes can be safely reverted.