aigit

Correlate AI coding sessions with git history.

aigit tracks which lines of your codebase were written by an AI assistant — and which weren't. It works by matching code blocks from AI session logs to git commits via fuzzy hashing, then surfacing that attribution as a git blame-style overlay and CI report.

   4 a1b2c3d  [claude_c 100%] def greet(name: str) -> str:
   5 a1b2c3d  [claude_c 100%]     return f"Hello, {name}!"
   6 a1b2c3d
   7 f9e8d7c                   def farewell(name: str) -> str:
   8 f9e8d7c                       return f"Goodbye, {name}!"

Why

As AI-assisted coding becomes the norm, teams lose visibility into:

• What percentage of the codebase is AI-authored
• Whether AI-generated code has higher churn or bug rates than hand-written code
• Which specific commits introduced AI-generated code and from what prompt

Install

pip install aigit

Requires Python 3.10+ and git.

Quick start

# 1. Index your AI sessions and git history
aigit index

# 2. View blame with AI attribution overlay
aigit blame src/myfile.py

# 3. See repo-wide AI% stats
aigit stats

# 4. Generate a CI report
aigit ci --format=markdown

Supported AI tools

The provider system is pluggable — see CONTRIBUTING.md to add your tool.

Commands

aigit index

Reads AI sessions and walks git history to build a local attribution database at .aigit/attribution.db.

aigit index                          # auto-discover Claude Code sessions
aigit index --sessions ./my-exports  # point to a custom session directory
aigit index --provider codex         # use a different provider

aigit blame <file>

Annotates git blame output with AI attribution. Lines written by an AI are highlighted with the tool name and confidence score.

aigit blame src/app.py
aigit blame src/app.py --show-prompt   # also show the prompt that generated each line

aigit stats

Shows AI attribution statistics across the repository.

aigit stats                    # repo-wide summary
aigit stats --file src/app.py  # single file
aigit stats --top 50           # show top 50 files (default: 20)

Example output:

aigit stats  —  3 sessions, 142 code blocks indexed

File                              AI lines  Total   AI%    Tools
────────────────────────────────  ────────  ──────  ─────  ─────────────
src/api/routes.py                      87     120  72.5%  claude_code:87
src/core/engine.py                     54      98  55.1%  claude_code:54
tests/test_routes.py                   41      60  68.3%  claude_code:41

Repo-wide: 182 / 450 lines (40.4% AI-attributed)

aigit ci

Emits a machine-readable report for CI pipelines.

aigit ci                        # markdown (default)
aigit ci --format=json          # JSON
aigit ci --output=report.md     # write to file

CI integration (GitHub Actions)

Add this workflow to post an AI attribution report on every pull request:

# .github/workflows/aigit-report.yml
name: aigit AI Attribution Report
on:
  pull_request:
    types: [opened, synchronize]
jobs:
  aigit-report:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - run: pip install aigit
      - run: aigit index
      - run: aigit ci --format=markdown --output=aigit-report.md
      - uses: marocchino/sticky-pull-request-comment@v2
        with:
          path: aigit-report.md
          header: aigit-attribution

How it works

1. Session ingestion — reads AI tool session logs and extracts code blocks (both markdown fences and file write operations)
2. Tiered fuzzy matching — compares code blocks against git diff hunks using:
   • Exact SHA-256 match → 100% confidence
   • TLSH fuzzy hash distance < 30 → 90% confidence
   • TLSH fuzzy hash distance < 100 → 70% confidence
3. Attribution storage — stores matches in a local SQLite database (.aigit/attribution.db)
4. Blame overlay — runs git blame and annotates lines where the originating commit matched an AI code block

Contributing

See CONTRIBUTING.md.

License

Apache 2.0 — see LICENSE.