nemo-relay-export-atif-trajectories

par nvidia

Export NeMo Relay activity as ATIF trajectories for replay, analysis, or interchange

npx skills add https://github.com/nvidia/nemo-relay --skill nemo-relay-export-atif-trajectories

Export ATIF Trajectories

Use this skill when the user wants execution traces as ATIF documents rather than live OTLP spans.

Default Path

  • Create an AtifExporter with session and agent metadata
  • Register it before the instrumented work
  • Run scoped tool and LLM activity
  • Call export() or export_json()
  • Clear between runs and deregister when done

Embedded ATIF Semantics

  • ATIF export translates NeMo Relay events into ATIF v1.7 trajectory data.
  • LLM start events become user steps; message content is extracted from the LLMRequest.content payload when possible.
  • LLM end events become agent steps with response content, model metadata, token metrics, reasoning fields, and promoted tool_calls when the response uses a supported tool-call shape.
  • Tool start events are skipped in the trajectory because tool calls are promoted from the preceding LLM end response.
  • Tool end events become system observations. Observations are correlated to promoted tool calls by function name and source call ID when available.
  • Mark events with data become system steps. Scope start/end events are structural and are not emitted as trajectory steps.
  • Scope nesting becomes ancestry metadata on exported steps.
  • Nested agent scopes become embedded subagent_trajectories with subagent_trajectory_ref observations in the parent trajectory.
  • Event payloads become step input, step output, tool-call content, or observation content.
  • The exporter preserves collected event order and uses lifecycle pairing to reconstruct the trajectory.
  • Exporting does not clear the buffer. Use one exporter per run or call clear() between runs when concurrent agents share a process.
  • Before using a trajectory in evaluation, confirm schema_version is ATIF-v1.7, agent metadata is correct, expected LLM/tool steps are present, tool observations follow tool calls, and sensitive fields are absent.

Important Semantics

  • ATIF exports the full event buffer collected so far.
  • Consecutive tool observations can be merged into one system observation step.
  • Trajectories reflect sanitized event payloads, not raw secrets that sanitize guardrails removed before event emission.
  • Response codecs can improve LLM end annotations, but they do not change the caller-visible LLM response.

Checklist

  • Session and agent metadata chosen
  • Exporter registered before the relevant run
  • Scope boundaries are correct so ancestry is meaningful
  • Export timing is clear: whole buffer vs clear-between-runs
  • LLM responses include tool_calls if ATIF tool-call entries are expected

Related Skills

  • nemo-relay-setup-observability
  • nemo-relay-instrument-calls
  • nemo-relay-debug-runtime-integration