Runs Python unit tests with coverage, analyzes coverage reports, and implements meaningful tests to increase coverage by ~0.2%. Use when you want to systematically improve Python test coverage with high-value test cases.
Increase Python unit test coverage by ~0.2% through meaningful tests that add real value.
Be fully autonomous — Do NOT stop or pause to ask for confirmation. Keep iterating (analyze → implement → verify) until the 0.2% coverage target is reached. If you encounter ambiguities about what to test, make a reasonable choice and proceed.
Step 1: Run tests with coverage
make python-tests # ~3 min, creates .coverage file
Generate JSON report for analysis:
uv run coverage json -o coverage.json
The JSON contains per-file missing_lines arrays showing uncovered line numbers.
Step 2: Analyze and prioritize
Read coverage.json to find files with:
percent_covered (high impact)lib/streamlit/elements/ or lib/streamlit/runtime/Skip: >97% coverage, proto/*, vendor/*, static/*, test files.
Step 3: Implement tests (in subagent)
Launch a subagent to implement tests for each prioritized file. Provide the subagent with:
missing_lines from coverageThe subagent should:
lib/tests/streamlit/<path>/<module>_test.pylib/tests/AGENTS.md: prefer pytest-style standalone functions over unittest.TestCase classes, use @pytest.mark.parametrize to consolidate tests that only differ in inputs/expected outputs, add numpydoc docstrings and type annotationsuv run pytest lib/tests/streamlit/path/to/module_test.py -vStep 4: Verify and iterate
uv run pytest lib/tests/streamlit/path/to/module_test.py -v # Run new tests
make python-tests # Measure progress
Repeat steps 2-4 until coverage improves by ≥0.2%, then run make check.
Step 5: Simplify, review, and address feedback
Once all tests pass and coverage target is met:
simplifying-local-changes subagent to clean up and simplify the code changes. Wait for completion.reviewing-local-changes subagent to review the changes. Wait for completion and read the review output.DO test: Conditional logic, error handling, edge cases (None, empty, zero, max), public API functions, complex branches.
DON'T test: Simple accessors, protobufs, implementation details, already well-covered code.
Coverage exclusions: Use # pragma: no cover sparingly for code that genuinely doesn't need testing. Always include a reason (e.g., # pragma: no cover - defensive):
# pragma: no cover - platform-specific)# pragma: no cover - defensive)# pragma: no cover - abstract)Integration dependencies: Packages listed under [dependency-groups] integration in pyproject.toml (e.g., pydantic, sympy, polars, sqlalchemy) are only installed for integration tests, not regular unit tests. When writing tests that use these packages:
@pytest.mark.require_integration marker to the testlib/tests/streamlit/<package>/<module>_test.py mirrors lib/streamlit/<package>/<module>.py
Quality > coverage numbers - skip tests that don't catch real bugs
Target is 95%+ coverage per lib/tests/AGENTS.md
Use /checking-changes after implementing tests
Some code paths involving external libraries (e.g., database connectors, optional dependencies) are already covered by integration tests marked with pytest.mark.require_integration. These integration tests are not included in the coverage numbers from make python-tests. When analyzing missing lines, check whether the uncovered code is exercised by integration tests before adding unit tests or # pragma: no cover annotations.
Local vs CI coverage differences: Code that is version-specific (Python version, library version) or uses integration dependencies may appear uncovered locally but is tested and covered in CI. Examples:
if sys.version_info >= (3, 14)) run only on matching CI jobs@pytest.mark.require_integration) run in separate CI jobs with those packages installedBefore adding tests or # pragma: no cover for such code, verify whether it's already exercised in CI.