refactor: add binary instruction content checklist

[FL4TLiN3]

Summary

• Replace subjective self-check with six binary pass/fail rules for instruction content quality
• Make instruction quality rules domain-agnostic (remove coding-specific examples)
• Update verify-test Step 3 to use same binary criteria as CONTINUE triggers

Why

The previous self-check ("would the LLM produce worse results?") was subjective — the LLM always judges its own output as necessary. Binary checks (code block present? library name present?) have clear yes/no answers with specific remediation actions, matching the pattern of structural checks (#2-6) that already work.

Checklist

1. No code blocks
2. No library/tool names (capability requirements instead)
3. No file paths
4. No procedures
5. No technique explanations
6. ≤15 line budget (non-coordinator)

Supersedes #773.

Test plan

• Run create-expert and verify generated instructions pass all 6 binary checks
• Verify structural checks still work (delegates, pick, modelTier, verifier)

🤖 Generated with Claude Code