Slopwatch: LLM Anti-Cheat for .NET

Purpose

A .NET tool designed to detect "reward hacking" behaviors where AI coding assistants take shortcuts instead of properly solving problems.

Detection Capabilities

• Disabled tests (Skip, Ignore, #if false)
• Warning suppressions via pragma or SuppressMessage
• Empty catch blocks swallowing exceptions
• Arbitrary delays masking timing issues (Task.Delay, Thread.Sleep)
• Project-level warning suppression patterns
• Central Package Management bypasses

Integration

• Runs as Claude Code hook during AI-assisted coding
• Integrates with CI/CD pipelines (GitHub Actions, Azure DevOps)
• Baseline tracking to catch only new issues
• JSON output for programmatic consumption
• Configurable suppression rules via .slopwatch/config.json

Workflow

1. Initialize baseline from existing code (slopwatch init)
2. Commit baseline to repository
3. Run analysis on new changes (slopwatch analyze)
4. Update baseline when intentional code is added (slopwatch analyze --update-baseline)

Key Quote

"When LLMs generate code, they sometimes take shortcuts that make tests pass or builds succeed without actually solving the underlying problem."

Apache 2.0 licensed.