Use when running deep performance and throughput investigations for HUD publishing paths (widgets/zones/tiles), including MCP HTTP benchmarks and gRPC bidi stream benchmarks. The canonical gRPC widget publish-load benchmark is `examples/widget_publish_load_harness` (Rust); the Python `grpc_widget_publish_perf.py` script is a secondary alternative.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Use when running deep performance and throughput investigations for HUD publishing paths (widgets/zones/tiles), including MCP HTTP benchmarks and gRPC bidi stream benchmarks. The canonical gRPC widget publish-load benchmark is `examples/widget_publish_load_harness` (Rust); the Python `grpc_widget_publish_perf.py` script is a secondary alternative.
User Test Performance
Run focused performance drills for publish throughput, latency, and transport bottlenecks.
This skill is separate from /user-test functional validation and is tuned for repeatable measurement.
Skill-Creator + Brainstorming Contract
Before any run, define and document:
Question: what exact performance hypothesis are we testing?
Target: which target_id are we measuring?
Workload identity: what is the benchmark primary key?
Success criteria: what threshold/regression signal matters?
This skill is built to make those answers machine-auditable and historically comparable.
Start with one target (user-test-windows-tailnet, same host as /user-test),
then add more (for example, a remote MacBook target) under new target_id keys.
Scripts
scripts/mcp_publish_perf.py
Benchmarks MCP publishes for widget or zone modes.
Uses local scripts/proto_gen/ stubs (self-contained inside this skill).
scripts/widget_soak_runner.py
Runs the Rust gRPC widget harness concurrently for agent-alpha, agent-beta, and
agent-gamma by default.
Defaults to a 60-minute paced soak (--duration-s 3600) and writes per-agent
artifacts plus soak_summary.json under benchmarks/soak/<timestamp>/.
Use with the benchmark Windows config (app/tze_hud_app/config/benchmark.toml)
and benchmark scheduled task (scripts/windows/install_benchmark_hud_task.ps1).
scripts/compare_results.py
Compares candidate vs baseline runs from reference/results.csv.
Reports metric deltas and threshold pass/fail for regression gates.
Run Selection (Progressive Discovery)
Use the minimum run shape that answers the current hypothesis:
Transport bottleneck hypothesis (per-request overhead, HTTP connection churn) -> mcp_publish_perf.py
The installer stores the PSK as a DPAPI-protected file for the task user and the
runner passes it to tze_hud.exe through TZE_HUD_PSK. It only stops an
existing benchmark-config tze_hud.exe process before relaunching; it does not
kill the production TzeHudOverlay process by executable name.
These fields are persisted in results.csv for auditable historical comparisons.
The Rust harness (examples/widget_publish_load_harness) uses structured thresholds via
--target-p99-rtt-us and --target-throughput-rps, with traceability embedded in the
emitted JSON artifact (RFC-0005 / publish-load-harness spec ID) and Layer 4 artifact
output (via --layer4-output-root).
Notes
MCP runtime path in this repo is currently one-request-per-connection (no keep-alive), so high-rate streams are transport-limited.
gRPC byte stats are protobuf payload bytes (ByteSize) and not full wire bytes with transport framing.
