Detect which test runner a project uses, without running any of its tests.

Looks for pytest (config, conftest.py, dependencies), unittest, jest and vitest (package.json, config files). For each candidate: the evidence, the exact way it would be run, and whether it is ready (pytest importable, node_modules present). For Python it also reports which interpreter would be used (project .venv, else the server's own) and the versions of pytest/coverage found there. Returns test directories too.

path: project root (absolute is safest). python: optional interpreter path, or 'uv'.

Run the tests and return counts plus only the useful details of each failure.

EXECUTES THE PROJECT'S CODE. Result: outcome (passed | failed | no_tests | timeout | crashed | startup_error), summary counts (passed/failed/error/skipped/xfailed/xpassed), failures (id, message, file:line of the innermost project frame, the relevant traceback frames only, trimmed stdout/stderr), collection_errors listed separately from failures (import errors never hide as test failures), the 5 slowest tests, and which interpreter was used. On timeout or crash it names the test that was running. The whole process tree is killed at the timeout.

selector: a file, directory or node id (pytest: tests/test_a.py::TestX::test_y; unittest: dotted name; jest/vitest: file path pattern). Default: everything. timeout: seconds, 1-3600. extra_args: only allow-listed flags, e.g. ["-x"], ["-k", "login"], ["-m", "not slow"], ["--maxfail=3"]; anything else is refused. python: interpreter path or 'uv'. runner: force pytest/unittest/jest/vitest (default: auto-detected, see detect_runner).

Re-run only the tests that failed or errored in the last run_tests/rerun_failed for this project (same server session). Returns the same structure as run_tests, plus how many ids were rerun (at most 100). The failure list is replaced by what still fails.

List the tests without running them (collect-only). Returns ids (max 500) and any collection errors (import errors in test modules or conftest.py), reported separately.

filter: pytest -k expression / unittest name pattern / substring of a jest or vitest file name. For jest and vitest only test FILES can be listed. Use the ids as run_tests selectors and as flaky_check / explain_failure test_id.

Run ONE test n times (2-20, each a fresh process) and report the pass/fail pattern.

Returns pattern like "PPFPP" (P pass, F fail, S skipped, T timeout), counts, a verdict (flaky | stable_pass | consistently_failing | inconclusive), the failure rate, and the first failure's details. Worst case it takes n x timeout seconds. A test that is only flaky when run with others (ordering, parallelism, shared state) can pass here; check that with run_tests on the surrounding file.

test_id: exact id from run_tests/list_tests (pytest: path::Class::test[param]; jest/vitest: path/file.test.js::full test name; unittest: dotted name).

Full detail for one failing test: re-runs just that test with the long traceback and (for pytest) --showlocals, and returns the complete traceback text (capped at 20k chars) with local variable values, the assertion message, and captured stdout/stderr. If the test passes when run alone, it says so (flaky or order-dependent).

Run the Python tests under coverage.py and report per-file percent and the uncovered line ranges (e.g. "12-15, 20"). Needs coverage installed in the project's interpreter (the error says which). Data goes to a temp dir; nothing is written into the project.

files: optional list of paths or globs relative to the project (src/app.py, src/*.py) to report; default is the 30 least-covered project files plus the total percent. selector: run only part of the suite. Python projects only.