slopo
The missing piece in AI‑assisted development
Open-source CLI tool for detecting non-exact code duplication separated across the codebase. A kind of problem frequently introduced by AI agents and difficult for them to detect.
Integration with coding agents
The core of the solution is Slopo CLI. Integration with agents is only an additional way of using it. But a powerful one.
With dedicated instructions (skills), agents run the Slopo CLI, analyze results, and guide a developer, providing the right level of detail depending on the situation. They support discarding nonactionable duplication, making decisions, and refactoring.
The following videos demonstrate how Claude Code and Codex review uncommitted changes, where Slopo reported a new implementation similar to an existing one.
How detection works
Slopo uses embedding models to identify similar logic implemented in different ways. These AI models are typically used to find texts that are phrased differently but have similar meanings. Some models are able to understand source code, and this tool uses this technology for duplicate detection.
Additionally, the distance between similar code in the codebase influences the final score. The further apart the code is, the higher its priority. These duplicates are the hardest to detect, both for humans and AI.
The wider picture
I built Slopo to solve a real, but narrow problem. Solutions like this are only part of something broader and are rarely enough on their own.
If your team, project, or product needs someone who sees the whole picture, see what I offer.
Where Slopo helps
Review
It finds similar code in recent Git changes, helping review the code. It analyzes changes in the context of the entire codebase, rather than looking only at the changed code in isolation. It can detect when AI-generated code implements logic that already exists elsewhere in the codebase.
Refactoring
It finds similar code across the entire codebase, supporting refactoring and maintenance.
Why agents need help here
Source of duplication
Without rigor, a clear architecture design, and a specification available to it, an AI agent tends to implement logic that already exists elsewhere. It isn't a copy of the code, it's a new implementation of the same solution.
Why it hurts
Duplicated code confuses AI, making it less reliable when deciding where to add new related logic or which existing code to edit. It may update one copy while the others drift out of sync. This problem isn't unique to AI, but AI creates chaos faster than humans do.
The blind spot
AI agents struggle with one specific kind of duplicate. They can find shared names, trace how code is called and referenced, or reason about the part of the system they are working in. But they can't load the whole codebase into context at once, so they miss distant functions that do the same thing in a different form.
Documentation
Full documentation is on GitHub, with installation, configuration options, and detailed usage.