Meta-skill that improves other skills based on real usage failures. When an SDT
reports a skill produced incorrect or unexpected output, this skill analyzes the
root cause, generates a structured improvement proposal, applies the fix following
CONTRIBUTING.md guidelines, and offers to create a PR or apply locally.
Also maintains the project learnings layer (distill mode): dedupe, retire
falsified entries, promote proven learnings into references/upstream.
Use when: "improve", "this didn't work", "fix this skill", "the output was wrong",
"skill improvement", "qa-improve", "distill learnings", "clean up learnings".
Do NOT use when: giving content feedback on output (use opt…
