Sentrux — Architectural Feedback Sensor for AI Agents

What it is

Sentrux is a real-time architectural sensor written in pure Rust that helps AI coding agents close the feedback loop for code quality improvement. It monitors your codebase for architectural patterns, complexity metrics, and code smells, then feeds structured feedback back to AI agents so they can iteratively improve their own output.

Sentrux targets teams using AI agents in production who want to ensure code quality does not degrade over time. It works with Claude Code, Cursor, and any agent that accepts structured feedback.

How it saves time or tokens

Without Sentrux, AI agents operate without visibility into the architectural quality of the code they generate. They may introduce high cyclomatic complexity, tight coupling, or naming inconsistencies that accumulate over time. Sentrux provides continuous monitoring with structured JSON output that agents can parse and act on, creating a self-improving loop. The estimated overhead is 2,400 tokens per analysis cycle.

How to use

1. Install Sentrux:

cargo install sentrux

1. Run watch mode for continuous monitoring:

sentrux watch ./src --rules complexity,coupling,naming

1. Integrate with your AI agent:

# Agent writes code, then Sentrux analyzes it
sentrux analyze ./src --format json | your-agent

# Example output:
# {"file": "src/auth/handler.rs", "line": 45,
#  "severity": "warn",
#  "rule": "complexity",
#  "message": "Cyclomatic complexity 23 (threshold: 15)"}

Example

The agent feedback loop in practice:

# Step 1: Agent writes code
codex --auto 'Implement the payment processing module'

# Step 2: Sentrux analyzes the result
sentrux analyze ./src/payments --format json > feedback.json

# Step 3: Agent reads feedback and improves
codex --auto 'Fix the issues in feedback.json:
  - Reduce complexity in process_payment() from 23 to under 15
  - Extract helper functions for validation logic
  - Fix naming: use snake_case consistently'

# Step 4: Verify improvements
sentrux analyze ./src/payments --format pretty
# [OK] All checks pass

Related on TokRepo

• Coding tools — More code quality and developer tools on TokRepo.
• AI tools for testing — Browse testing and validation tools.

Common pitfalls

• Setting complexity thresholds too low generates excessive warnings that agents cannot prioritize. Start with standard thresholds and tighten gradually.
• Running Sentrux on generated code (tests, migrations) produces false positives. Exclude auto-generated directories from the analysis path.
• Relying solely on Sentrux for code review misses semantic issues that static analysis cannot catch. Use Sentrux as a complement to human code review, not a replacement.