adit-code

[](https://github.com/justinstimatze/adit-code/actions/workflows/ci.yml) [](https://goreportcard.com/report/github.com/justinstimatze/adit-code) [](https://opensource.org/licenses/MIT)

Most of your code is fine. But in every codebase there are a handful of files where your AI coding agent burns disproportionate tool calls — chasing imports across files, re-reading large modules it can't hold in context, wading through ambiguous grep results. adit finds those hot spots and tells you exactly what to fix.

Validated against 8,964 real Claude Code sessions: adit's strongest metric — single-consumer imports that should be co-located — correlates at r=+0.735 with actual AI tool call counts. Across 33 open source projects, 84-88% of files score Grade A across 33 open source projects. The few percent that don't account for the majority of AI navigation cost.

> *An adit is a horizontal tunnel driven into a hillside to provide access > to a mine. adit-code tunnels into your codebase to find the structural hot > spots that make AI editing expensive.*

AI coding tools (Claude Code, Codex, Cursor, aider) navigate code by grepping and reading files. Every import from another file is a Read tool call. Every ambiguous function name is a grep that returns noise. Every large file requires multiple reads where the AI loses coherence across distant methods.

Most files are small and well-structured — the problems concentrate in a few hot spots. But those hot spots are where the AI spends most of its time. adit identifies them so you can fix the files that matter most.

Metrics correlated against actual AI tool call counts from production sessions (8,964 Claude Code sessions across Python and TypeSc