Agent eval suite: 25 tests guard output quality

What's new

Two new test files run on every pull request:

• -`test_agent_compliance.py`: 11 tests covering frontmatter schema, model selection, communication protocol, identity lines, severity indicators, and emoji usage
• -`test_skill_compliance.py`: 14 tests covering output-kit contract, atlas-report overflow clause, naming conventions, and structured workflow steps

Drift tracking uses explicit allowlists: known violations are enumerated and tests fail when drift is fixed but not removed from the allowlist, so nothing slips through unreviewed.

Also fixed

12 skills were missing the output-kit contract reference line. Fixed in: draft-landing, draft-patterns, form-exam, form-palette, form-style, lens-chart, pitch-landing, prism-chart, prism-stack, proof-design, surge-landing, touch-ui.