[FEATURE] LLM Code Review Assistant — AI-powered code review and suggestions

[gelluisaac]

Description

Implement an LLM-powered code review system that automatically reviews pull
requests, suggests improvements, and catches potential issues.

Scope

Build intelligent code review assistant.

Files to Touch/Create

• astroml/llm/code_review/__init__.py
• astroml/llm/code_review/reviewer.py — Code review logic
• astroml/llm/code_review/analyzers/ — Language-specific analyzers
• astroml/llm/code_review/checks/ — Review check implementations
• astroml/llm/code_review/suggestions.py — Improvement suggestions
• .github/workflows/llm-code-review.yml — PR review workflow

Review Capabilities

1. Security: SQL injection, XSS, auth issues
2. Performance: Inefficient algorithms, memory leaks
3. Style: PEP8, best practices
4. Correctness: Logic errors, edge cases
5. Testing: Missing tests, weak assertions
6. Documentation: Missing docstrings
7. Complexity: Overly complex functions

Implementation Details

• Analyze diffs, not entire files
• Context-aware: understand surrounding code
• Categorize issues by severity
• Suggest specific fixes with code snippets
• Learn from human reviewer decisions
• Integrate with GitHub PR comments

Acceptance Criteria

• Catches >70% of issues human reviewers find
• False positive rate <20%
• Review completes within 2min of PR creation
• Suggestions are actionable
• Learns from accepted/rejected suggestions
• Supports Python, SQL, YAML

Review Format

## Security
- **[HIGH]** SQL injection risk in `query.py:45`
  - Use parameterized query instead of f-string

## Performance
- **[MEDIUM]** N+1 query in `accounts.py:123`
  - Consider using `joinedload`

## Style
- **[LOW]** Consider extracting magic number to constant

Labels

enhancement, llm, code-review, developer-tools