testgap 0.1.0a0

AI-powered test generator that closes coverage gaps in your PRs — only suggests tests that actually pass.

TestGap

AI-powered test generator that closes coverage gaps in your PRs — only suggests tests that actually pass.

Status: alpha. testgap init and testgap diff work today (non-interactive). testgap diff --review is in progress.

TestGap analyzes the uncovered code in your diff, generates pytest tests using your chosen LLM (Claude, GPT, Gemini, or local Ollama), and only suggests tests that have been executed and passed. Provider-agnostic, diff-focused, and verified — no hallucinated tests.

Install

pip install -e ".[llm]"   # llm extra pulls in litellm for live calls

(Once published: pip install testgap[llm].)

Quick start

cd your-project
testgap init                  # detects pytest / src layout / test dir, writes .testgap.yml
testgap diff                  # generate tests for uncovered diff lines, validate them
testgap diff --base origin/main --max-functions 3

testgap diff runs end-to-end:

1. Resolves the base ref (origin/HEAD → main → master, override with --base).
2. Computes the changed-line set from git diff base..HEAD.
3. Runs pytest --cov to find which changed lines are not executed.
4. Groups uncovered lines into enclosing functions/methods via AST.
5. For each function, calls the configured LLM with a few-shot prompt that imitates existing tests in the project.
6. Writes the generated test to a temp file under tests/, runs pytest against it, and only reports tests that actually passed.

Suggestions are not committed automatically — that's the --review interactive mode (coming next).

Configuration

testgap init creates a .testgap.yml like:

version: 1
project:
  language: python
  test_framework: pytest
  source_paths: ["src/"]
  test_paths: ["tests/"]
coverage:
  threshold: 80
  diff_threshold: 90
  exclude:
    - "**/migrations/**"
    - "**/__init__.py"
llm:
  model: anthropic/claude-sonnet-4-6
  max_cost_per_run: 2.0
  max_retries: 2
generation:
  style: match_existing
  include_docstrings: true
  max_tests_per_function: 3

LLM provider

TestGap uses LiteLLM for provider abstraction. Set the matching env var for the model you choose:

Model	Env var
anthropic/claude-sonnet-4-6	ANTHROPIC_API_KEY
openai/gpt-4o	OPENAI_API_KEY
gemini/gemini-2.0-flash	GEMINI_API_KEY
ollama/qwen2.5-coder	— (local)

Development

pip install -e ".[dev]"
pytest

License

MIT