What problem does it solve? When a Summer skill fails behavioral spec assertions or produces low-quality output, reading the skill text alone often cannot reveal why. This Skill runs the skill in parallel subagents with and without proposed changes, scores both versions against the spec, and ships the winner. ## Core Features & Use Cases - Baseline Evaluation: Spawns subagents with the current skill body against each spec Case, capturing tool calls, diffs, and assertion scores. - Parallel A/B Comparison: Repeats the run with a proposed SKILL.md revision and compares assertion pass rates, tool-call efficiency, and hallucination penalties per case. - Guarded Shipping: Overwrites the skill and commits only after explicit user confirmation, with a structured win/tie/loss summary. - Use Case: A contributor's skill fails 2 of 6 spec assertions. Run this Skill to baseline the failures, draft a fix, verify the fix wins on net across all cases, and ship it with a conventional commit. ## Quick Start Ask the agent to run /skill-improve on the skill named <name> using the cases in tests/specs/<name>.md and compare the proposed revision against the baseline.
