Workspace Smoke Testing

Use this as the router. For workspace-specific expectations, read the matching <workspace>-workspace-testing skill first, then run the shared scripts under tests/workspace_smoke/.

Choose Scope

• Explicit workspace: pass --workspace <name>_ws.
• Changed files: pass --changed-from <base> --changed-to HEAD, or pipe paths with --changed-files-from -.
• Shared testing changes under skills/, tests/workspace_smoke/, docker_modules/, or setup scripts select all workspaces.

If the user has not chosen the validation depth and the run is expensive or local GUI/GPU dependent, ask whether to rebuild images and whether GUI proof is needed. If they only need a quick non-GUI check, prefer image-cli.

Test Levels

• Build: python3 tests/workspace_smoke/run.py --workspace <ws> --level build
• CLI without Compose startup: python3 tests/workspace_smoke/run.py --workspace <ws> --level image-cli
• Runtime CLI through Compose: python3 tests/workspace_smoke/run.py --workspace <ws> --level cli
• Isaac headless renderer proof: python3 tests/workspace_smoke/run.py --workspace <ws> --level isaac-visual
• Isaac Lab deformable smoke: python3 tests/workspace_smoke/run.py --workspace <ws> --level isaac-lab-deformable
• Documented GUI/Isaac proof: python3 tests/workspace_smoke/doc_demo_smoke.py --workspace <ws> --summary-json tests/workspace_smoke/artifacts/<ws>-doc-demo.json
• Go2 rough-terrain training and inference proof: python3 tests/workspace_smoke/go2_rough_smoke.py --workspace go2_ws

For a new image, build locally and refuse registry pulls with docker compose up -d --build --pull never. After a successful build, pass --no-build to run.py or doc_demo_smoke.py to reuse that image without rebuilding it. The Go2 rough-terrain runner reuses a local image by default and accepts --build when needed. The standalone Isaac GUI capture and deformable helpers also accept --no-build. Keep direct docker run only for cheap image CLI and Docker GPU preflight checks.

For documented Isaac joint-command motion proofs, capture the whole sequence: open the stage with the timeline paused, validate the expected robot prim, start host X11 recording, trigger Isaac timeline Play, wait for ROS topics and scene stabilization, then publish the documented /joint_command. The recording must include the start of simulation and pre-command frames, not only the post-command motion.

For Isaac Lab deformable-object Kit GUI proof, run tests/workspace_smoke/isaac_lab_deformable_gui.sh --detach, follow the printed log path, and wait for Registered backend 'kit', [INFO]: Setup complete..., and the first Root position (in world) line. Setup complete can still precede the first rendered viewport frame. Always take and inspect a check screenshot before recording; if it shows a black viewport, wait until the orange deformable objects render, then record. Confirm MP4 proof by extracting a frame from the recording and verifying it shows the deformable scene. The Kit GUI tutorial must run with a TTY; use docker compose exec from an interactive terminal, or script -qefc ... /dev/null from non-interactive smoke runners. For automated checks, poll the log inside the Compose service rather than the outer PTY transcript; the host transcript can lag or remain attached after the container has already emitted root-position lines. Expect first Compose-based Kit launches to take minutes while Isaac/RTX/cache startup work completes; compare Simulation App Starting to Simulation App Startup Complete before calling it a deformable simulation hang.

For screenshot/recording proof requests, prefer the direct GUI capture path: launch the workspace GUI, wait for a visible readiness marker/window, then use gui-proof-capture / proof_capture.py to capture the host X11 display. Treat isaac-visual as a deeper renderer smoke test, not the default proof path; use it only when the user specifically asks for a headless render, deterministic rendered scene, or simulator image output.

For Isaac GUI proof scenes, run tests/workspace_smoke/isaac_gui_proof_scene.py inside the workspace container and wait for ISAAC_GUI_SCENE_READY before calling proof_capture.py. The helper authors a cube, ground, lights, and camera view, then keeps Isaac open for host-side screenshot or recording capture.

Use --no-gpu only for config/build/image-cli checks. Use --disable-gpu-reservation only as an explicit fallback when Compose runtime checks are needed on a host with broken NVIDIA container runtime state.

Proof Reports

Write JSON summaries with --summary-json. Convert saved manifests and repository test logs into Markdown with:

python3 tests/workspace_smoke/report.py \
  --summary smoke=tests/workspace_smoke/artifacts/<summary>.json \
  --test-log test-all=tests/workspace_smoke/artifacts/<test-log>.log \
  --output tests/workspace_smoke/artifacts/proof-report.md

For GUI screenshots or recordings, use the shared gui-proof-capture skill and tests/workspace_smoke/proof_capture.py. GUI proof captures the host X11 desktop/window after the container renders to the host display.

Reusable Test Logic

Treat tests/workspace_smoke/artifacts/ as the normal output location for logs, screenshots, recordings, and temporary proof data. Avoid /tmp for validation work because a crash or reboot can erase it; use /tmp only when a tool requires it, and immediately copy any useful result back into the repository's artifact tree. Do not leave reusable validation scripts in output-only or ephemeral locations.

When a docker-module or workspace validation needs a repeatable helper, promote it to a git-tracked path before finishing:

• General repo helpers: tests/workspace_smoke/
• Shared Python utility code: tests/workspace_smoke/lib/
• Docker-module-specific helpers: tests/workspace_smoke/modules/<module>/
• Workspace-specific helpers: tests/workspace_smoke/workspaces/<workspace>/

If a prompt teaches a reusable workflow rule, update the relevant skill in skills/. Skills should stay concise: point to the tracked script, state readiness markers, capture/report expectations, and avoid embedding large script bodies.

Guardrails

• Keep heavy Docker/GPU checks out of tests/test_all.sh.
• Run scripts/post_install.sh when Docker module hard links may be stale.
• Use --continue-on-failure for broad proof collection.
• Always include log and artifact paths in the final report.
• Before finishing, check whether any ad hoc command or script should become a tracked helper or skill instruction for future reuse.
• Run ./tests/test_all.sh after changing scripts, skills, workflows, or linted repository structure.