test-coverage-improver

Run the coverage suite with make coverage, then inspect the console output of coverage report -m for file-level totals and missing line numbers. For interactive drill-downs, generate an HTML report with uv run coverage html and open htmlcov/index.html.

Prioritize public APIs and shared utilities in src/agents/ over examples or docs, then target files with low statement coverage or newly added code at 0%. Focus on risky paths like error handling, retries, timeouts, and concurrency rather than trivial happy paths.

No, it presents a numbered list of proposed test additions with expected coverage gains and explicitly asks for approval before editing any code or fixtures. Implementation only begins after the user agrees.

Rerun the coverage command to regenerate .coverage data and coverage.xml instead of guessing from outdated results. Stale artifacts can misrepresent current coverage and lead to poorly targeted test proposals.

Tests that hit uncovered paths such as error cases, boundary inputs, optional flags, and cancellation or timeout behavior yield the largest gains. Covering combinational logic beats adding more happy-path tests, and async tests should avoid flaky timing dependencies.