sentinel

Sentinel — Unified Code Review Engine

What It Does

Sentinel combines the best of 7 code review approaches into one unified system:

• Multi-agent parallel review (from code-review plugin)
• Adaptive confidence scoring (novel — learns your signal/noise preference)
• Security pattern blocking (from security-guidance, expanded to 16 patterns)
• Risk classification and churn detection (from clamper)
• Specialized domain agents (from pr-review-toolkit)
• Non-git support (novel — tracks edits via PostToolUse hooks)

How To Use

Slash Commands

• /sentinel:review — Full 6-agent review (quality, bugs, errors, security, tests, polish)
• /sentinel:review quick — Fast 2-agent check (quality + bugs)
• /sentinel:review security — Security-focused review
• /sentinel:review tests — Test coverage analysis
• /sentinel:review quality — Quality + simplification
• /sentinel:review src/file.py — Review specific files
• /sentinel:quick-check — Alias for quick mode

Auto-Detection

Sentinel automatically detects what to review:

1. Git projects: Uses git diff for changed files
2. Non-git projects: Uses the edit log (tracked by PostToolUse hook) to find files modified this session
3. Explicit files: Pass file paths as arguments

Adaptive Thresholds

The confidence threshold starts at 80 and adapts:

• If you say "show more" or "lower threshold" → threshold decreases (more findings)
• If you dismiss findings or say "too noisy" → threshold increases (fewer findings)
• Security scanner starts at 70 (more sensitive by default)
• Each agent has its own threshold

Agents

Agent	Model	Focus
sentinel-reviewer	Sonnet	CLAUDE.md compliance, conventions, code quality
bug-hunter	Opus	Logic errors, null handling, race conditions, edge cases
error-auditor	Sonnet	Silent failures, empty catches, broad exceptions
security-scanner	Opus	OWASP Top 10, data flow analysis, trust boundaries
test-analyzer	Haiku	Behavioral test coverage, critical gaps, test quality
code-polisher	Haiku	Post-review simplification preserving behavior

Hooks (Always Active)

• PreToolUse: Blocks edits containing security vulnerability patterns (16 patterns)
• PostToolUse: Tracks all edits for scope detection and risk classification

Data Files

• edit-log.jsonl — Edit tracking log (the non-git diff)
• sentinel-config.json — Thresholds, agent config, review modes
• security-patterns.json — Extensible security pattern library