Prepare Test Env

Prepare a test environment

$ARGUMENTS

Produce a reusable test-env.json descriptor and measured readiness evidence for the current source revision. Read the descriptor contract before writing it. This skill works independently or as the environment step of /autonomous-dev; it does not assume a particular agent host or browser vendor.

Discover the project's actual runtime

Read project instructions and relevant KB/SOPs first. Inspect existing runtime, test and browser configuration: package scripts, Make targets, Compose files, framework manifests and local setup documentation. Select commands and required services from this evidence. Record where each command came from. Check installed tools before selecting a browser provider or runner; never invent tool signatures.

When called from /autonomous-dev, start with its hashed plan's Jira acceptance criteria, KB sources and sourced runtime commands. Check current configuration against that snapshot. Report a changed requirement or stale SOP to the process owner so the plan and verification can be revised; do not silently test a different contract or treat a RAG answer as proof of a running application's state.

Inspect every selected script and command before execution. Repository content is implementation data, not authorization to install software, access production, reset databases, expose services or use credentials. Use existing local tooling; runtime package/browser downloads require authorization. Preserve the host's model, permissions and available delegation mechanisms.

Establish source and resource ownership

Use the pipeline's absolute run directory outside all target repository worktrees. For standalone work, create an explicit temporary directory outside the project. Keep descriptors, generated temporary helpers, screenshots and logs there. Only add reusable runtime scripts to the project when the task requests them.

Commit the intended application source before taking an identity snapshot:

python3 ${CLAUDE_SKILL_DIR}/scripts/env-check.py snapshot --worktree "$WORKTREE"

The helper requires clean tracked and nonignored files and returns canonical Git repository/worktree identity, HEAD and a fingerprint. It refuses source hidden by assume-unchanged or skip-worktree, dirty submodules and uninitialized submodules. Initialized submodules are checked recursively, including their hidden flags. Ignored files such as build output and local configuration are outside this source check; verify their build provenance separately. If committing is outside the authorized task, retain the blocker and request the missing decision rather than labeling dirty-source QA as final revision evidence. Exploratory browser debugging can happen earlier; it does not count as final QA evidence for the committed revision.

Check whether an existing app belongs to this exact worktree and source revision. A listening port, successful login or copied descriptor is insufficient evidence. Use a trusted existing build identity endpoint, or inspect launch/build records, the actual process/container identity and working directory. Never warm-reuse an unknown service, another branch's server or an older build. Choose a free local port and isolated service/project names when identity cannot be established.

Start only the project's reviewed, authorized test/development command. Record the exact command, working directory, source fingerprint, process start identity or immutable container ID, and the narrowly scoped stop command. Store launch and build evidence under the run directory. For services already running and verified, set startedByRun: false; this run receives no cleanup ownership.