---

name: grpc-and-developer-tools description: "Maintain AlpaSim protobuf and gRPC contracts, use generated stubs, inspect installed packages and entry points, and perform safe contributor/tooling checks without starting simulation or external infrastructure." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

gRPC and developer tools

Use this sub-skill when the task is about the AlpaSim protobuf contract, generated Python modules, service-client examples, package metadata, plugin entry points, contributor checks, map-tool boundaries, telemetry inspection, or deciding whether a helper is safe to run. This is a developer/API skill, not a simulation operator.

Route first

• For installation, Hydra composition, local/Docker/Slurm deployment, scene caches, or run telemetry operation, use simulation-wizard.
• For service lifecycle, runtime orchestration, replay, timing, or service implementation, use runtime-services.
• For model/plugin behavior or driver configuration, use drivers-and-plugins.
• For ASL inspection, metrics, map/trajectory analysis, or result evaluation, use evaluation-and-logs.

Fast workflow

1. Identify the installed alpasim_grpc version and the exact .proto or service involved. Do not infer a message name from a nearby service.
2. For a schema change, preserve package names, field numbers, and import paths; regenerate all Python artifacts using the package build/compile workflow. Never hand-edit generated *_pb2.py, *_pb2_grpc.py, or .pyi files.
3. Inspect the diff, import the generated modules, and run a focused native test plus formatting/static checks before proposing a service change.
4. For a plugin or package change, inspect its declared entry-point group, import target, dependency direction, and optional backend requirements.
5. Keep scheduler submission, SSHFS mounts, Docker launchers, model downloads, and full simulations as explicit user-approved operations. This sub-skill only supplies safe inspection and generation guidance.

Detailed API facts and service grouping are in references/grpc-api.md. Read tools and safety before a helper; the protobuf compiler is the bundled non-destructive path for explicit source/output roots. The contributor sequence and review checklist are in references/developer-workflows.md. Use references/tools-and-safety.md before running helper commands and references/troubleshooting.md for recovery.

Generated API use

The public generated package is alpasim_grpc.v0. Typical client imports are:

import grpc
from alpasim_grpc.v0 import common_pb2, sensorsim_pb2
from alpasim_grpc.v0.sensorsim_pb2_grpc import SensorsimServiceStub

with grpc.insecure_channel("host:port") as channel:
    stub = SensorsimServiceStub(channel)
    request = sensorsim_pb2.RGBRenderRequest(scene_id="scene_id")
    # response = stub.render_rgb(request, timeout=10)

Use the actual generated message and stub names; for example, the render message is RGBRenderRequest, not a generic RenderRequest. A channel/client smoke test proves serialization and stub construction, not that a renderer or other service is running. Route service implementation and health diagnosis to runtime-services.

Commands and gates

• In an installed package, use uv run compile-protos and uv run clean-protos only from the gRPC package's own project context. The bundled helper at scripts/compile_protos.py is for explicit source and output paths and is non-destructive by default.
• For a proto edit, compile, inspect generated-file changes, run generated imports, then run the narrowest relevant test. A successful compile does not establish wire compatibility.
• Run uv run pytest <focused-test> for safe unit cases and pre-commit run --all-files for contributor-facing changes when the repository environment is available. Do not substitute a broad simulation run for these checks.
• alpasim-info reports installed model, MPC, scorer, tool, and config entry points. It may enumerate optional model imports; treat a timeout or logged load warning as an optional-dependency issue, not proof that the registry is empty.

Boundaries and handoff

This skill documents map visualization and telemetry/scheduler helper safety, but does not launch them. A map helper may require an artifact and optional geometry/plotting packages; use a disposable output location and route actual map interpretation to evaluation-and-logs. Prometheus/Grafana and Slurm helpers can mount, chmod, submit, requeue, or create files outside the project; read their help and obtain explicit approval first. Never copy credentials, model weights, scene caches, SSH paths, or scheduler allocations into a skill.

When handing off, report: schema/package touched, generated files checked, commands and tests run, optional dependencies skipped, and whether the operation contacted a service, scheduler, network, Docker, or filesystem mount.