| name | three-dtiles-stream-tokyo-tower-live |
| description | Run real Resonite Link `stream` verification around Tokyo Tower and compare the current repo against the latest `v*` tag. Use when the user wants machine-level live-send or performance regression checks instead of dry-run or unit tests, including session discovery, stale-session cleanup, host-side Windows execution, log capture, summary extraction, `visibleMs` aggregation, and latest-tag comparisons. |
3D Tiles Stream Tokyo Tower Live
Use this skill for real stream validation against a Windows-hosted Resonite Link session. Prefer unit tests and dry-run checks first. Switch to this skill when the question is about live send ordering, log-backed refinement/removal behavior, wall-clock regressions, or comparison against the latest release tag.
Warning: this workflow removes existing 3DTilesLink Session ... roots in the target live world. Use it only in a disposable session, or after the user has clearly accepted that cleanup.
Workflow
-
Resolve the listener port, then verify it from the same Windows host context that will run dotnet.exe.
Prefer an explicit --host and --port when the target listener is already known. If auto-discovery is used, treat the discovered endpoint as provisional until the script's host-side TCP preflight passes. The script tries localhost first and then the discovered announce address as a fallback host candidate.
-
Verify protocol responsiveness before cleanup or stream.
The script sends a minimal host-side send-json request after TCP connect succeeds. Treat TCP success without protocol response as invalid.
-
Run the standard Tokyo Tower live case with scripts/run_tokyo_tower_stream.py.
Use it when you need one live run with logs, summary extraction, visibleMs stats, the host-side TCP preflight result, and the host-side protocol preflight result.
-
Use scripts/cleanup_sessions.py when you need an explicit packaged cleanup step.
-
Run scripts/compare_latest_vtag.py when the question is regression-oriented.
It prepares a worktree for the latest v* tag, builds both trees on Windows, runs the same live scenario with --no-build, and prints a JSON comparison.
-
Treat preflight failure, stderr, missing summary lines, or non-zero FailedTiles as invalid.
If UDP discovery succeeds but the script reports host-side TCP preflight failure or host-side protocol preflight failure, do not continue with cleanup or performance conclusions. Re-resolve the port or fix the listener first.
-
If the user says the listener is already up, verify that claim with the host-side preflights instead of assuming that UDP announcements are current.
-
Use the reference doc for interpretation details.
Do not call the run clean if the app did not emit the final CandidateTiles=... ProcessedTiles=... StreamedMeshes=... FailedTiles=... line.
Read references/workflow.md when you need the expected markers, common failure modes, or how to judge visual/performance drift.
Scripts
-
scripts/run_tokyo_tower_stream.py
Run one live Tokyo Tower stream case, clean the live world first, capture stdout/stderr logs, and emit a JSON summary.
The script first verifies that the selected host/port accepts a TCP connection from a Windows host process, then sends a minimal send-json request from the same host context before cleanup or stream.
-
scripts/cleanup_sessions.py
Run the packaged cleanup step for stale 3DTilesLink Session ... roots on the selected live session.
-
scripts/compare_latest_vtag.py
Compare the current repo against the latest v* tag with the same live Tokyo Tower scenario and emit JSON with per-case summaries plus deltas.
Output
Return the script JSON plus the key facts:
- listener port
- host-side TCP preflight result
- host-side protocol preflight result
- stdout/stderr log paths
- wall-clock duration
CandidateTiles, ProcessedTiles, StreamedMeshes, FailedTilesvisibleMs count, median, p95, max- whether the latest-tag comparison changed output counts or latency
