Add review depth modes (quick/default/thorough)

[lionuncle]

Summary

• Adds --depth/-d flag with three review modes:
  • quick — blockers only: production bugs, security vulnerabilities, data loss risks, major architectural violations. Skips nits, style, and minor edge cases. Ideal for fast approvals on low-risk PRs.
  • default — standard review behavior (unchanged from today).
  • thorough — deep analysis: architecture fit, edge cases, naming clarity, API design, testability, performance at scale. For critical code paths and complex changes.
• Depth instructions are injected into the system prompt so they shape the entire review, not just a checklist.
• Status line shows the active depth when non-default (e.g. 🤖 Reviewing with claude-opus-4-6 [quick]...).

Usage

rpr 42 --depth quick       # or -d quick
rpr 42 --depth thorough    # or -d thorough
rpr 42                     # default (unchanged)

Test plan

• rpr <pr> -d quick --dry-run — verify review only flags critical issues, approves clean code
• rpr <pr> -d thorough --dry-run — verify review covers architecture, naming, scale concerns
• rpr <pr> --dry-run — verify default behavior is unchanged
• rpr <pr> -d invalid — verify argparse rejects unknown depths

🤖 Generated with Claude Code