ux(ar): per-round task tree for FULL mode round 2+

ux(ar): replace TodoWrite with TaskCreate/TaskUpdate for E→A→J visibility

feat(run): integrate autotune health check as optional pre-grind step

feat(gates): prevent padding — hard discard if skill inflates >50% without new examples

feat(metrics): add padding_alarm — detect inflation attacks in real experiments

docs(metrics): document attack vectors for disabled skill_depth/agent_sections

feat(metrics): replace gameable metrics with structural proxies (anti-Goodhart)

fix(prompt-testing): correct test directory and unit test patterns

feat(prompt-testing): scan-and-scaffold.sh — repo-wide gap detection with recency priority

refactor(prompt-testing): scaffold-test.sh + inline helpers + script-first execution

fix(test-helpers): cross-platform comment (macOS + Linux, no GNU timeout)

fix(prompt-testing): cheat mode enforcement + execution section + superpowers reference

chore(release): bump to v0.5.1 — Xavier School Migration & Skill Polish

feat(xavier-school): migrate skills, agents, hooks, and tests into autoimprove

docs(research): trim autoresearch threat memo to 442 words (≤500 limit)