Verify a change in developerz-ai/claude-task-master

House skill — established by dz-runner setup mode from this repo’s own detected commands. Correct it in place; setup never overwrites a skill that exists.

The one gate

bin/check is the local gate, also invoked by the pre-commit hook. CI does not invoke it: .github/workflows/ci.yml independently runs equivalent lint, test, and PR-package commands. Keep this script aligned with those gates; a local pass is evidence for the local environment, not a guarantee that CI will pass.

Supported setup and startup

bin/setup supports Debian/Ubuntu-style hosts with Bash, dpkg and apt-get. Install mise and Docker with the Compose plugin first, and put mise-managed tools on PATH (including uv after mise use). Compose must support up --wait and --wait-timeout. The caller needs Docker daemon access and, when apt packages are missing, root privileges or working non-interactive sudo -n access. Network access is needed for apt, mise and uv downloads. Setup retains the global mise selections for uv 0.13 and Python 3.14 and installs all project extras. It creates claudetm-profiles only after dependency sync succeeds; it never reads, copies or provisions credentials.

bin/dev builds from the checked-out Compose build context rather than relying on a potentially stale image. Compose waits for the existing REST /health check; startup/health failures return nonzero, with a 180-second readiness wait bound. This is not a wall-clock bound on the image build. Successful startup leaves the stack detached. Compose mounts and credential isolation are unchanged: health readiness does not require a personal login or an agent credential, and is not proof that authenticated agent operations work.

uv run pytest tests/test_bin_scripts.py -q --timeout=120 runs copied scripts in a temporary repo using PATH-injected fake tools and an isolated HOME. It pins ordering, root/sudo package installation, fail-fast behavior and dev rebuild/wait arguments without invoking real apt, Docker, credentials or global mise changes. These regressions are not evidence of real downloads, image builds or service health. The coordinator's read-only survey could not execute real setup/dev; record actual bin/setup, bin/dev and bin/check results separately when run on a supported host. Do not claim real startup based on fake-tool tests alone.

CI-to-local gate mapping

Use Python 3.14 and install all extras before proving a change: uv sync --all-extras --python 3.14 (also performed by bin/setup). CI selects Python 3.14 and installs the editable package with uv pip install -e ".[all]".

Local tests deliberately use two workers rather than CI's -n auto to stay within the shared box's resource ceiling. TMPDIR=/tmp remains scoped to pytest. Coverage produces .coverage data, coverage.xml, and a terminal missing-lines report. These artifacts are local evidence; the script does not upload them.

The build uses a unique temporary directory under ${TMPDIR:-/tmp} and cleans it on exit, including failed builds or smoke checks. Pre-existing dist/ artifacts are neither overwritten nor deleted. As in CI, the CLI probes run against the editable project environment, not an installation of the built wheel.

Lint, formatter check, typecheck, tests, build, and CLI help must all succeed; the script stops on failure. Only doctor and status use CI's explicit || true tolerance: missing credentials or project state may make them fail. Their output remains visible, but a successful gate does not certify those two probes passed.

CI's test-output/build-artifact uploads and Codecov submission are reporting, not local gates. Tag-triggered package/Docker publishing and manual test publishing workflows are release operations, not part of this local PR gate. No workflow restructuring is needed to keep these commands aligned.

Rules

• Run every command from the repo root — the scripts cd there themselves.
• bin/check red is not a pull request. Fix the cause, never silence the gate.

Exit codes bin/check must use

A coding box runs this gate before it commits, and it has to tell a FAILING change from a gate it could not run — otherwise correct work is thrown away as if it were red. Exit 1 cannot carry that difference, so one code is reserved:

• Guard preconditions FIRST and exit 75 when one is unmet. exit 1 there tells the platform your change is broken when it is your environment that is.
• Changed a dependency? Re-run bin/setup before bin/check, or you are proving a stale tree.
• Add the test that fails first, then the fix — a green run that never went red proves nothing.