Skip to content
Codegen Sandbox Codegen Sandbox

run_failing_tests

Rerun only the tests that failed in the most recent run_tests call in this session. Skips anything that passed, so the fix-rerun loop drops from minutes to seconds on large codebases.

The sandbox composes a narrow test-runner invocation from the structured failure records it already captured for last_test_failures. After the rerun completes, the structured-failure store is overwritten with the fresh results — a follow-up last_test_failures reflects the new state.

Schema

Section titled “Schema”
ParamTypeRequiredDefault
limitnumberno50 (clamped to 200) — max distinct test names in the rerun filter
timeoutnumberno300s (max 1800s) — same semantics as run_tests

Behaviour

Section titled “Behaviour”
  1. If no run_tests call has been made yet in this session → text result no run_tests call yet in this session.
  2. If no supported project is detected in the workspace root → error result.
  3. If the detector has no structured-failures parser (v1: every language except Go) → text result run_failing_tests: <language> detector has no structured failures; rerun run_tests manually.
  4. If the last run had zero failures → text result last run_tests had no failures — nothing to rerun (<language>).
  5. Otherwise the sandbox composes a Go -run regex from the stored failures and executes it. Output shape matches run_tests (stdout, optional stderr section, optional timeout line, trailing exit: N).

Language support

Section titled “Language support”
  • Go — composes go test -json -count=1 -run '^(TestA|TestB|...)$' <packages>. Packages are the set of import paths from the failure records; when the set is ≤ 10 distinct packages they’re passed positionally, otherwise the invocation falls back to ./... to keep argv bounded. Subtest failures rerun the parent test (Go’s -run regex matches the top-level name).
  • Node / Python / Rust — parser not yet wired. The tool surfaces a not supported for <language> notice; rerun run_tests directly.

Examples

Section titled “Examples”

After a failing run:

{"tool": "run_tests"}
→ 3 failures: pkg/a:TestAlpha, pkg/a:TestBeta, pkg/b:TestGamma

Rerun only those:

{"tool": "run_failing_tests"}

The composed invocation is go test -json -count=1 -run '^(TestAlpha|TestBeta|TestGamma)$' pkg/a pkg/b.

Limit the filter (useful when dozens of tests failed and you want to focus on the first batch):

{"tool": "run_failing_tests", "arguments": {"limit": 5}}

Store semantics

Section titled “Store semantics”
  • Reads the same TestResultStore slot last_test_failures reads from.
  • Writes the slot on completion, so a follow-up last_test_failures surfaces the rerun’s results — not the original run’s.
  • Session-scoped. Process-local; no persistence across restarts.
Section titled “Related”