SNAFU Review 2026: An LLM flow that fixes bad names in your code

What SNAFU Does

SNAFU (Symbol Name Ambiguity Fixer-Upper) is an agentic LLM flow for the classic hard problem: naming things. For each symbol in a file, it computes a Name Ambiguity Number (NAN) — a measurable score for how ambiguous the name is — then walks you through a pipeline to replace it with a clearer name, with a human confirming the real meaning along the way.

Key Features

• Quantifies name quality via Shannon entropy / perplexity (NAN)
• Human-in-the-loop confirms the real meaning before any rename
• Drops non-improving candidates (NAN delta ≤ 0) and sanity-checks survivors
• Multi-language via Tree-sitter: Python, Ruby, C#, Java, JS, TS, PHP, Rust, Go
• Any-LLM support (LiteLLM-style) with a simple CLI

Who Should Use SNAFU

Engineers cleaning up legacy or AI-generated code where names like process_data or handle hide intent. Useful in code review prep and onboarding hygiene.

Pros and Cons

Pros

• Turns “name quality” into a measurable number
• Conservative: human confirms meaning, weak renames dropped
• Broad language support via Tree-sitter

Cons

• Requires Python 3.14+ and an LLM API key
• Name scoring is an LLM heuristic, not ground truth
• Not all symbols extracted — focuses on the most relevant

Pricing

Free and open source under MIT.

FAQ

How is ambiguity measured?

Each symbol is scored with no context; the model returns interpretation probabilities, and NAN = 2 ** Shannon entropy (perplexity). Lower is clearer.

Which languages are supported?

Python natively, plus Tree-sitter extraction for Ruby, C#, Java, JavaScript, TypeScript, PHP, Rust, and Go.