agentleFS
NVIDIAKnown publisherNVIDIA/NeMo-Relay/.agents/skills/maintain-dynamic-plugins/SKILL.md

maintain-dynamic-plugins

Skill238 starsChanged 2 months agoApache-2.0View on GitHub

Change NeMo Relay dynamic plugin loaders, manifests, native ABI, gRPC worker protocol, or Python worker SDK. Do not use for ordinary built-in plugin configuration.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this skill for plugin.kind = "rust_dynamic", plugin.kind = "worker", nemo-relay-plugin, nemo-relay-worker, nemo-relay-worker-proto, …

    line 9
  2. Very specific:

    Keep the stable boundary explicit: native plugins cross a C ABI.

    line 15
  3. Distinctive:

    Do not pass Rust runtime types, trait objects, futures, or allocator-owned strings across the native dynamic-library boundary.

    line 17
  4. Very specific:

    Keep relay-plugin.toml dynamic records separate from generic runtime components.

    line 24
  5. Very specific:

    Treat compat.relay as the plugin author's compatibility assertion, not proof that an artifact was rebuilt.

    line 30
  6. Very specific:

    Do not add tests under src.

    line 38

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Maintain Dynamic Plugins

Use this skill for plugin.kind = "rust_dynamic", plugin.kind = "worker", nemo-relay-plugin, nemo-relay-worker, nemo-relay-worker-proto, nemo-relay-types, and the Python nemo-relay-plugin package.

Rules

  • Keep the stable boundary explicit: native plugins cross a C ABI; worker plugins cross nvidia.nemo.relay.worker.v1 (contract 1.0.0).
  • Do not pass Rust runtime types, trait objects, futures, or allocator-owned strings across the native dynamic-library boundary.
  • Typed native middleware futures run on the SDK-owned Tokio executor. Keep subscribers synchronous and preserve raw synchronous ABI registrations.
  • Define closed worker transport structures in protobuf when generated clients must enforce their fields. Keep open application payloads lossless by using JsonValue or JsonObject rather than google.protobuf.Value.
  • Keep relay-plugin.toml dynamic records separate from generic runtime components. Enabled records can create internal component specs. Disabled records remain available for inspection but do not load.
  • Require migrated worker plugins to regenerate their bindings, rebuild, and declare a compat.relay range covering the new contract. Recommend >=1.0.0,<2.0 for the current worker SDK.
  • Treat compat.relay as the plugin author's compatibility assertion, not proof that an artifact was rebuilt. Do not add a legacy raw-result adapter.
  • The worker protocol is nvidia.nemo.relay.worker.v1, contract version 1.0.0. The native host uses ABI v8 and accepts v7 callback layouts. Guard append-only extensions with struct_size. Keep existing fields at their original offsets. A breaking native JSON contract needs a new native_api value. A breaking C layout needs a new ABI version. A breaking worker contract needs a new protocol identity.
  • Do not add tests under src; Rust tests belong in crate tests/ trees and Python SDK tests belong under python/tests.
  • Native and worker plugins are trusted extensions. Document that native plugins are in-process and unsandboxed; worker plugins provide process isolation but not a security sandbox.

Checklist

  • Manifest validation covers kind, compatibility, load contract, integrity, capability mismatch, and disabled-plugin behavior.
  • Native loader keeps libraries alive until registered callbacks are cleared and deregisters plugin kinds before unload.
  • Worker activation covers process launch, token auth, handshake, validation, declarative registration, proxy rollback, cancellation, and shutdown.
  • Rust and Python SDKs expose every supported registration surface.
  • Runtime helpers cover marks, scopes, continuations, and isolated scope stacks.
  • plugins list, plugins inspect, and plugins validate report lifecycle and compatibility status without leaking secret config.
  • Top-level doctor reports resolved dynamic plugin and host configuration status.
  • When detailed dynamic plugin guides exist, they keep Rust native, Python worker, and nvidia.nemo.relay.worker.v1 protocol details on separate pages.
  • justfile, Codecov, and CI package/test workflows include new plugin crates and packages.

Validation

Choose checks by the changed layer:

  • Manifest or shared types: test the owning Rust crate.
  • Native loader or ABI: prepare plugin fixtures and run the focused native plugin integration test.
  • Worker protocol or host: test the worker crates and regenerate or test the Python worker SDK only when its protocol surface changes.
  • Shared runtime behavior: run the Rust suite and only binding suites whose observable plugin behavior changes.
  • Documentation or packaging: run their targeted checks only when changed.

Canonical surface suites prepare plugin fixtures. Before a raw focused native or worker integration test, run just build-test-plugin-fixtures; never compile fixtures inside an individual test case.

References

  • crates/plugin-host/src/dynamic/
  • crates/plugin
  • crates/worker
  • crates/worker-proto
  • crates/types
  • python/plugin
  • examples/rust-native-plugin
  • docs/build-plugins
  • examples/python-grpc-worker-plugin

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.