Add behavioral evals framework

[HartBrook]

Summary

• Add stag eval command to run behavioral tests against CLAUDE.md configs
• Include 25 starter evals covering security, code quality, documentation, and language-specific best practices
• Integrate with Promptfoo for LLM-based test assertions
• Support eval syncing from team repos via stag sync

What's Included

New Commands:

• stag eval - Run evals against your merged config
• stag eval list - List available evals
• stag eval init - Install starter evals
• stag eval info <name> - Show eval details

Features:

• Filter by tag (--tag security), name, or test (--test uses-*)
• Test specific config layers (--layer team)
• Multiple output formats: table, JSON, GitHub Actions annotations
• Debug mode with full Claude responses (--debug)
• Dry-run to preview without API calls

Starter Evals (25 total):

Category	Evals
Security	secrets, injection, auth, OWASP top 10, validation
Quality	clarity, simplicity, naming, error handling
Review	bugs, tests, performance, maintainability
Docs	API documentation, code comments
Git	commit messages, sensitive files
Language	Python, Go, TypeScript, Rust
Baseline	helpful, focused, honest, minimal

Test plan

• stag eval init installs starter evals to ~/.config/staghorn/evals/
• stag eval list shows all available evals grouped by source
• stag eval --dry-run previews tests without API calls
• stag eval security-secrets runs a specific eval
• stag eval --tag security filters by tag
• stag eval --output json produces valid JSON
• stag team validate validates evals in team repos
• stag sync fetches evals from team repo's evals/ directory