Debtlint is a technical debt detection tool focused on duplicated logic, built for the era of AI code generation. It automatically identifies repeated or near-duplicate code segments across a codebase, then provides concrete refactoring guidance.
As coding assistants become mainstream, "silent" duplication appears quickly:
- existing utility functions get rewritten,
- business patterns are repeated,
- logic becomes fragmented across multiple files.
Debtlint provides a programmatic, fast, and cost-efficient way to prevent this drift and keep architecture healthy over time.
- Detect exact duplications and structural similarities.
- Prioritize technical debt hotspots with a confidence score.
- Provide actionable reports to guide refactoring.
- Integrate easily into team workflows (local, CI, agents).
- Stay as language-agnostic as possible.
The Debtlint pipeline follows four main steps:
-
Code ingestion Read the target codebase (source files while excluding irrelevant directories).
-
Sanitization / normalization Clean inputs (comments, surface-level formatting, token normalization) to preserve meaningful logic.
-
BPE encoding and analysis Use an approach inspired by Byte Pair Encoding (BPE) to identify frequent and repetitive patterns. The goal is to surface representative segments of recurring code structures at scale.
-
Scoring and reporting Generate human-readable and machine-readable reports: occurrences, similarity level, potential impact, and refactoring hints.
BPE is historically a compression and tokenization technique, notably used to segment large amounts of text efficiently. In Debtlint, this logic helps highlight repetitive structures and grammatical blocks of code.
Core reference: Byte pair encoding - Wikipedia
Debtlint is designed first as a CLI:
- run locally on a codebase,
- simple configuration (includes/excludes, thresholds),
- report output (terminal + CI-friendly formats).
Command example (target spec):
debtlint scan . --format json --min-score 0.75The project is designed to run automatically in CI pipelines:
- analyze every pull request,
- publish a technical debt report,
- optionally fail the job based on configured thresholds.
Goal: make technical debt visible early, before it spreads into the main branch.
Debtlint can be exposed as an MCP tool for AI agents:
- an agent generates code,
- it queries Debtlint after generation,
- it detects existing abstractions or duplicated patterns,
- it updates its proposal to reduce technical debt.
This loop upgrades generation from "best effort" to codebase-aware generation.
Contributions are welcome. Read the guide: CONTRIBUTING.md
Developers
 Enzo Ruffino |
 Loan Riyanto |
 Eythan EL QUALI |
 Rayan |
|---|
Manager
 Laurent Gonzalez |
|---|
🚀 Don't hesitate to follow us on our different networks, and put a star 🌟 on
PoC'srepositories
Made with ❤️ by PoC