refactor(evaluate-plugin): extract evaluate skill inline commands to standalone scripts

[laurigates]

Problem

Two skills in evaluate-plugin use 4-5 shell utility Bash patterns (jq, cat, mkdir, wc, date) in their allowed-tools frontmatter. This forces Claude to generate inline bash commands that each require individual user approval, since complex/compound commands can't be allowlisted via Bash(command *) patterns.

Affected skills

Skill	Shell utility patterns	Count
evaluate-skill	jq, cat, mkdir, wc, date	5
evaluate-plugin-batch	jq, cat, mkdir, date	4

Expected behavior

Consolidate shell utility operations into standalone scripts in each skill's scripts/ directory and use Bash(bash *) permission pattern. The evaluate plugin already has a scripts/ directory at the plugin root level — consider whether skill-level scripts or shared plugin-level scripts are more appropriate.

Reference

• See health-plugin/skills/health-check/ for the correct pattern
• Rule: .claude/rules/agentic-permissions.md → "Scripts for Compound Operations"
• Detected by: scripts/plugin-compliance-check.sh Check 6 (check_bash_patterns)