Documentation

Everything you need to understand how Audit.git scores your commits.

Overview

Audit.git uses a three-stage pipeline to evaluate every commit message: a deterministic rule engine, NLP feature extraction, and GPT-4o semantic analysis. The results are combined into a weighted composite score from 0 to 100.

// Composite score formula

final_score = (rule_score * 0.4) + (llm_score * 0.6)

API Reference

POST/analyze

Start analysis job for a public repo URL. Returns a job object immediately.

{ "repo_url": "https://github.com/user/repo", "max_commits": 100 }

GET/analyze/{job_id}

Poll job status. Status: pending | running | complete | failed.

GET/analyze/{job_id}/commits

Fetch scored commits with pagination. Query: skip, limit.

GET/analyze/{job_id}/export

Export results as CSV.

Grade Scale

A

80-100

Elite

B

60-79

Solid

C

40-59

Average

D

20-39

Weak

F

0-19

Chaos

Rule Engine

25pts

Length

Subject line between 10 and 72 characters.

15pts

Casing

Sentence case or Conventional Commits type prefix.

10pts

No Trailing Period

Subject line must not end with a period.

25pts

Imperative Verb

Message starts with an imperative verb (Add, Fix, Refactor, etc.).

15pts

No Generic Words

No "fix", "update", "wip", "stuff" without context.

5pts

Body Present

Multi-line commit with explanatory body.

5pts

Line Length

Body lines wrapped at 72 characters.