Code Review
Multi-dimensional code review based on Claude Code best practices.
Rule sources: CLAUDE.md (project instructions) + REVIEW.md (review-specific rules).
Review Modes
| Argument | Mode | Command |
|---|
(empty) or --staged | staged | git diff --cached |
--unstaged | unstaged | git diff |
--all | all | git diff HEAD |
--pr <number> | PR | gh pr diff <number> |
--commit <hash> | commit | git show <hash> |
--branch | branch | git diff main...HEAD |
--file <path> | file | Read file directly |
Review Dimensions
1. Bug Detection
- Logic errors, boundary conditions, null pointers, type errors
- Missing exception handling, resource leaks
- Concurrency issues, race conditions
- Hardcoded values, magic numbers
2. Security Review
- OWASP Top 10 vulnerabilities
- Missing input validation
- Sensitive data exposure (API keys, passwords, tokens)
- SQL/command injection risks
- Insecure dependencies
3. Architecture & Code Quality
- CLAUDE.md compliance
- REVIEW.md rule violations
- Code style consistency
- Naming conventions
- Function complexity (warn if cyclomatic > 10)
- Code duplication, over-engineering
4. Project-Specific (AlgVex) — from REVIEW.md
- Critical: Zero truncation, SL safety, layer order integrity, API key exposure, position sizing
- High: SSoT sync, Telegram Chinese display, agent data flow, R/R guarantees, NT API usage
- Medium: Config layer violations, error handling, async safety, feature extraction parity
- Nit: Code style, Occam's razor, documentation sync
Confidence Scoring
| Score | Meaning | Action |
|---|
| 0-49 | Possible false positive | Don't report |
| 50-79 | Medium confidence | List in "Suggestions" |
| 80-100 | High confidence | Must report |
Default threshold: >=80%
Output Format
# Code Review Report
## Summary
- Review scope: [mode description]
- Files: N
- High confidence issues: N
## Issues Found
### [Critical] Issue Title
- **File**: path/to/file.py:123
- **Confidence**: 95%
- **Type**: Bug | Security | Architecture | Project
- **REVIEW.md Rule**: #N (if applicable)
- **Description**: Detailed description
- **Suggestion**: Fix recommendation
## Suggestions (50-79% confidence)
- Issue list
## Conclusion
Review passed / Found N high-confidence issues
Severity Levels
| Level | Confidence | Action | REVIEW.md Mapping |
|---|
| Critical | >=90% | Block merge | Rules 1-5 |
| High | >=85% | Should fix | Rules 6-10 |
| Medium | >=80% | Recommend fix | Rules 11-14 |
| Low/Nit | >=70% | Optional | Rules 15-17 |
Key Files (Extra Scrutiny)
| File | Review Focus |
|---|
strategy/ai_strategy.py | SL/TP logic, layer orders, emergency paths |
strategy/order_execution.py | Bracket safety, trailing stop, position sizing |
strategy/safety_manager.py | Emergency SL retry, naked position detection |
strategy/event_handlers.py | Layer lookup, SL/TP pairing, position state |
agents/multi_agent_analyzer.py | Schema validation, data truncation, prompt injection |
agents/mechanical_decide.py | Threshold decision, zone conditions, direction lock |
utils/telegram_bot.py | side_to_cn() SSoT, message splitting, dual-channel |
main_live.py | Environment config, adapter setup |
configs/base.yaml | All business parameters (SSoT) |
Post-Review Validation
# After review, always run:
python3 scripts/smart_commit_analyzer.py # Regression detection
python3 scripts/check_logic_sync.py # SSoT sync check
