dsh-test-runner

When an agent runs tests through bash, it must read hundreds of lines of raw output token by token, inflating context and adding extra rounds. dsh-test-runner collapses that loop into a single structured tool call: it auto-detects the project test framework (vitest, jest, pytest, node:test, etc.), executes the tests, and returns only a pass/fail summary with failing case names, key error messages, and a tail of output — so the model can decide its next fix without wading through the full log.

Beyond auto-detection, the plugin supports file- or test-case-level targeting (e.g. tests/test_api.py::test_login), a one-flag rerun of only the previously failed cases, full command override (pnpm vitest run, python -m pytest -x -q), timeout enforcement, and sandbox-policy passthrough. The failure parser includes a graceful-degradation path: if a framework changes its output format, the tool does not crash — it simply falls back to exitCode plus output tail, no worse than running bash directly.

Built for developers assembling a code-change → test → fix automation loop in DeepSeek Harness. If your agent repeatedly runs tests, reads results, and pinpoints failures, this plugin compresses each test cycle into one clean JSON call, cutting token spend and interaction rounds.