feat(examples): add code-reviewer quickstart

[vadim-rl]

Summary

Add a new code-reviewer/ quickstart example that follows the same structure and conventions as the existing deep_research/ example. This gives users a ready-to-run agent for analyzing source code correctness, style, and security.

What's included

• code_reviewer/tools.py — Three custom deepagent tools:

  • analyze_code: AST-based static analysis (syntax, naming conventions, docstrings, function complexity, line length)
  • security_audit: Regex-based security scanner (eval/exec, pickle, SQL injection, shell injection, hardcoded secrets, unsafe yaml.load)
  • think_tool: Strategic reflection pattern (same as deep_research)

• code_reviewer/prompts.py — Three instruction sets mirroring the deep_research structure:

  • REVIEW_WORKFLOW_INSTRUCTIONS: 6-step review workflow
  • REVIEWER_INSTRUCTIONS: Sub-agent instructions with hard limits
  • SUBAGENT_DELEGATION_INSTRUCTIONS: Parallelization strategy

• agent.py — Standalone LangGraph agent definition with sub-agent support

• README.md — Self-contained with setup (uv sync), two usage options (direct Python invocation + LangGraph server), custom model/instructions/tools sections with tables

• pyproject.toml, langgraph.json, .env.example — following existing patterns

• Updated examples/README.md to list the new example

How to run

cd examples/code-reviewer
uv sync
export ANTHROPIC_API_KEY=your_key_here

Then either run via LangGraph server (langgraph dev) or use the direct Python invocation from the README.

Review notes

• All patterns (file structure, import style, docstring format, README layout, tool registration) follow the deep_research/ example exactly
• No existing files were modified except examples/README.md
• The tools use only Python standard library (ast, re) — no additional external dependencies beyond the deepagents stack
• Verified with unit testing: all tools produce correct output for edge cases (empty code, syntax errors, clean code, mixed security patterns)

This contribution was developed with the assistance of AI agents to generate code following existing project conventions.