giton 0.1.5

Local AI layer for git: orchestrates policies & plugins between commit and push.

giton

AI Cost Tracking

• 🤖 LLM usage: $0.6123 (8 commits)
• 👤 Human dev: ~$318 (3.2h @ $100/h, 30min dedup)

Generated on 2026-05-27 using openrouter/qwen/qwen3-coder-next

---

giton is a local AI layer for Git that works between commit and push. It helps standardize commits, propose safe code fixes, and orchestrate external tools via plugins.

Does something like this already exist?

Partially: there are tools for AI commit messages, git hooks, and commit-history rewriting. What is still missing is one local operator that:

• interacts with the user right after commit,
• proposes code fixes as additional commits (fixup!),
• cleans up history before push,
• integrates plugins via MCP/REST/CLI/gRPC.

Proposed direction for giton

1. Local-first with safe defaults

   • AI proposes changes, user approves them.
   • Prefer fixup! commits over automatic history rewriting.

2. Git hook layer

   • pre-commit: policy validation and quick fixes.
   • post-commit: inspect the fresh commit and propose follow-up patches.
   • pre-push: standardize history (autosquash, commit naming, final checks).

3. Plugin architecture

   • Shared input/output contract (JSON schema).
   • Plugin adapters: MCP, REST, CLI shell, gRPC/protobuf.

MVP usage example

After giton init, hooks are installed and run automatically from Git lifecycle events. The commands below show equivalent manual execution for demonstration/debugging.

# 1) initialize hooks in the repository
giton init

# 2) user makes a normal commit
git add -p
git commit -m "update stuff"

# 3) equivalent manual run: post-commit hook logic
giton hook post-commit

# 4) equivalent manual run: pre-push hook logic
giton hook pre-push

Example interaction:

giton: Found 2 issues (example: null check, commit message policy).
giton: Apply patch and add commit "fixup! ..."? [Y/n]

MVP plan

• MVP 1: hooks + policy engine + interactive CLI
• MVP 2: patching + fixup workflow + pre-push autosquash
• MVP 3: stable plugin API and MCP/REST/CLI/gRPC integrations

Install

pip install giton

Quick start

giton init                # install git hooks + 3 default plugins
giton shell               # interactive REPL
giton plugin catalog      # browse all available plugins
giton plugin install domd # install a specific extension
giton plugin install-category lang:python   # install everything for a language

Default plugins

The 3 plugins activated by giton init cover the most common day-to-day needs in a commit → push loop:

name	category	trigger	role
pyqual	lang:python	pre-commit	Python lint / type / complexity checks
vallm	task:validate	post-commit	Validate AI-generated code/patches
pretest	task:test	pre-push	Run / generate tests before push

Quick-extend categories

Each plugin in the catalog is tagged with a category, so you can install groups at once:

• languages: lang:python, lang:markdown, lang:any
• tasks: task:validate, task:test, task:refactor, task:autofix, task:fix, task:docs, task:security
• integrations: integration:mcp

giton plugin install-category task:autofix

Interactive shell

$ giton shell
giton> help
giton> install-defaults
giton> hook pre-commit
giton> catalog
giton> install prefact

Built-in policies

giton ships with a small zero-dependency policy engine that runs on every hook trigger — even before any plugin is installed:

• conventional_commits — subject must match type(scope)?: subject and stay within max_subject_length.
• no_wip_commits — blocks subjects matching wip, tmp, xxx, fixme.
• no_secrets — scans staged additions for AWS keys, private-key headers and api_key=/secret= patterns.
• max_file_size — rejects staged files larger than kb (default 512 KB).

Inspect or customize them per repo:

giton policy list                    # show active policies
giton policy check -t pre-commit     # evaluate without running plugins
giton policy init                    # write .giton/config.yaml

.giton/config.yaml is a deep-merge over the defaults — disable a single check or tweak max_subject_length without restating everything:

policies:
  no_wip_commits:
    enabled: false
  conventional_commits:
    max_subject_length: 100
hooks:
  post-commit:
    fail_on_policy: true   # turn advisory checks into blocking

Plugin contract

A plugin is any executable command (CLI). The catalog entry declares its trigger, category, install target (PyPI/local path) and command template. The runner expands {paths}, {diff_file} and {root} placeholders with the current git context before invocation.

Future exec types (mcp, rest) will share the same JSON in/out contract: input = git context + policy findings, output = list of proposed actions (patches, fixup commits, warnings).

Examples

The examples/ directory contains working demonstrations of giton:

• examples/basic - Basic usage example showing how to use giton as a Python library to collect git context, run triggers, and handle policy findings and plugin results. Can be run directly or with pytest.

• examples/advanced - Advanced usage example demonstrating plugin management (add, remove, list), custom policy configuration, hook installation/uninstallation, and running multiple triggers in sequence.

• examples/testing - Testing example with pytest integration and Docker support. Shows how to integrate giton into CI/CD pipelines with containerized testing environments. Includes Dockerfile and docker-compose.yml for easy setup.

Running examples with Docker

All examples include Dockerfiles for containerized execution:

# Basic example
docker build -f examples/basic/Dockerfile -t giton-example-basic .
docker run --rm -v $(pwd)/examples/basic/logs:/app/logs giton-example-basic

# Advanced example
docker build -f examples/advanced/Dockerfile -t giton-example-advanced .
docker run --rm -v $(pwd)/examples/advanced/logs:/app/logs giton-example-advanced

# Testing example
docker build -f examples/testing/Dockerfile -t giton-test .
docker run --rm -v $(pwd)/examples/testing/logs:/app/logs giton-test

See each example's README.md for detailed usage instructions.

License

Licensed under Apache-2.0.