agentleFS
NVIDIAKnown publisherNVIDIA/NeMo-Relay/.agents/skills/add-middleware/SKILL.md

add-middleware

Skill238 starsChanged 2 months agoApache-2.0View on GitHub

Add a new NeMo Relay guardrail or intercept type, registration surface, or pipeline stage. Do not use for changing an existing middleware implementation without a new middleware contract.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 1 is distinctive, and 8 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.

  • 8

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when introducing a new middleware registration surface or adding middleware behavior to a new pipeline stage.

    line 14
  2. Very specific:

    If this is an event sanitizer, which of data, category_profile, and metadata can change, and is the event used only as immutable context?

    line 30
  3. Very specific:

    Add the registry field to NemoRelayContextState in crates/core/src/api/runtime/state.rs.

    line 63
  4. Very specific:

    Add a SortedRegistry<GuardrailEntry<MyNewFn>> or SortedRegistry<Intercept<MyNewFn>> field to the state struct.

    line 66
  5. Very specific:

    Add registration and deregistration APIs in crates/core/src/api/.

    line 69
  6. Very specific:

    Use the existing global_*_registry_api!

    line 71
  7. Very specific:

    Add chain execution helpers to NemoRelayContextState in crates/core/src/api/runtime/state.rs.

    line 75
  8. Very specific:

    Follow the pattern of tool_sanitize_request_chain or tool_request_intercepts_chain.

    line 78
  9. Very specific:

    Update the relevant lifecycle owner to call the new chain method at the appropriate pipeline stage.

    line 82

≈ “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

Add a Middleware Type

NeMo Relay supports guardrails (validate/gate) and intercepts (transform) at various pipeline stages. Adding a new middleware type requires checking every layer that exposes the new contract.

Use this skill when introducing a new middleware registration surface or adding middleware behavior to a new pipeline stage.

Lock The Design First

Decide these before editing code:

  • Is this for tools, LLMs, marks, scope events, or a combination?
  • Is it a conditional guardrail, sanitize guardrail, request intercept, or execution intercept?
  • Does it run on request input, inner callable execution, stream chunks, or final response output?
  • Is the callback fallible, and how should callback failures propagate?
  • Does it need both global and scope-local registration?
  • What should subscribers and exporters observe in the event payload after this middleware runs?
  • If this is an event sanitizer, which of data, category_profile, and metadata can change, and is the event used only as immutable context?

Pipeline Order

Refer to docs/about-nemo-relay/concepts/middleware.mdx for the full diagrams.

  • Tool execute: conditional guardrails -> request intercepts -> sanitize request (for events) | execution intercept chain(callable) -> sanitize response
  • LLM execute: conditional guardrails -> request intercepts -> sanitize request (for events) | execution intercept chain(callable) -> sanitize response
  • Mark and scope events: specialized tool or LLM sanitizer (when applicable) -> mark or scope event sanitizer -> subscriber and exporter dispatch

Tool execution callbacks and each execution-intercept next continuation return the canonical ToolExecutionResult { result, annotation }. A forwarding intercept must preserve both fields in ToolExecutionInterceptOutcome; Relay retains pending_marks separately. Tool sanitize-response guardrails receive only result. Scope-end event sanitizers govern the annotation after Relay projects it to category_profile.tool_result_annotation.

Core Steps

  1. Define or reuse the callback type alias in crates/core/src/api/runtime/callbacks.rs.
pub type MyNewFn = Box<dyn Fn(&str, Json) -> Json + Send + Sync>;
  1. Add the registry field to NemoRelayContextState in crates/core/src/api/runtime/state.rs.

Add a SortedRegistry<GuardrailEntry<MyNewFn>> or SortedRegistry<Intercept<MyNewFn>> field to the state struct.

  1. Add registration and deregistration APIs in crates/core/src/api/.

Use the existing global_*_registry_api! and scope_*_registry_api! macro patterns in crates/core/src/api/registry.rs. Both global and scope-local variants are needed unless the design explicitly rules one out.

  1. Add chain execution helpers to NemoRelayContextState in crates/core/src/api/runtime/state.rs.

Follow the pattern of tool_sanitize_request_chain or tool_request_intercepts_chain.

  1. Wire the chain into the execute path.

Update the relevant lifecycle owner to call the new chain method at the appropriate pipeline stage. Tool and LLM paths live in crates/core/src/api/tool.rs and crates/core/src/api/llm.rs; shared mark and scope event sanitization lives in crates/core/src/api/shared.rs and is called from crates/core/src/api/scope.rs.

  1. Expose the new middleware surface in every affected binding.

For a public middleware contract, implement the Rust source of truth, then update only the bindings, FFI, wrappers, documentation, and tests that expose or observe the new contract.

Required Tests

  • Registration and duplicate-name behavior
  • Deregistration and no-op missing-name behavior
  • Ordering by priority
  • Callback failure policy, including fail-open behavior when required
  • Scope-local registration, inheritance, and cleanup on pop
  • Event payload semantics after middleware mutation
  • Tool execution result and annotation preservation, replacement, and removal when the middleware touches tool execution
  • Mark and scope event field semantics, including immutable identity fields
  • Parity coverage in every affected binding

Key References

  • Pipeline logic: crates/core/src/api/tool.rs, crates/core/src/api/llm.rs
  • Type aliases: crates/core/src/api/runtime/callbacks.rs
  • Runtime state and chain builders: crates/core/src/api/runtime/state.rs
  • Scope-local registry merging: crates/core/src/context/registries.rs
  • Registry: crates/core/src/registry.rs
  • Pipeline docs: docs/about-nemo-relay/concepts/middleware.mdx
  • Architecture docs: docs/about-nemo-relay/architecture.mdx
  • Registration examples: docs/instrument-applications/advanced-guide.mdx

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.