[VimalN2005]

Problem

In #731, Workflow.continue_run and Workflow.acontinue_run were instrumented to emit invoke_workflow spans. However, resolved human-in-the-loop (HITL) step requirements (StepRequirement) were omitted from gen_ai.input.messages because workflow steps are orchestration checkpoints rather than LLM tool calls (role: "tool"), as described in #906.

Solution

• Map resolved StepRequirement objects during Workflow.continue_run and Workflow.acontinue_run to gen_ai.input.messages with role: "user", representing human interaction and input that resumes the workflow.
• Handled attributes on resolved requirements:
  • step_id & step_name
  • confirmed, confirmation_message, rejection_feedback
  • user_input
  • selected_choices
  • output_reviewed and edited_output
• Respect content capture toggle: when content capture is disabled (OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=false), step requirements are omitted from spans.
• Follow Agno convention: when reading from run_response.step_requirements, only the active (latest) step requirement is captured.
• Added comprehensive unit tests and towncrier newsfragment.

Fixes #906

Type of Change

• New feature (non-breaking change which adds functionality)
• Tests

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

[opentelemetry-pr-dashboard]

Pull request dashboard status

Waiting on the author · refreshed 2026-11-01 22:38 UTC

Respond to 1 review item (e.g. link a commit, explain why not, ask a follow-up):

• Top-level threads: 1

Status above doesn't look right?

• Just replied or pushed? Anything around or after the refresh time above may not be picked up yet — give it a few minutes.
• Should this be with reviewers? Comment /dashboard route:reviewers to route it to them.
• Anything wrong — including the routing? Report it with what you expected; it helps us improve the dashboard.

[VimalN2005]

Thanks @DylanRussell! I reviewed semconv PR #188 — that proposal introduces individual step execution spans (run_step operation with gen_ai.step.name and duration metrics), but explicitly deferred step-level events and input.messages reshaping to a follow-up.

Workflow.continue_run and Workflow.acontinue_run instrument the workflow invocation span itself (invoke_workflow). Representing the human interaction/decision that resumes the workflow as a role: "user" input message on invoke_workflow provides immediate observability into the human inputs driving the continued run, while complying strictly with existing GenAI semconv for invocation spans. Once PR #188 lands and individual step spans are added here, having continuation inputs on invoke_workflow remains complementary. That said, if you'd prefer to wait or adjust the representation, happy to follow your guidance!

[lmolkova]

Thanks for picking this up! I'm not sure gen_ai.input.messages is the right place for this as is. The user message currently carries a JSON dump of Agno state (step_id, step_name, executor ids, tool_call_id, confirmation_message), and most of that isn't what the human said. confirmation_message is the developer-defined prompt shown to the human, and executor-level results are tool responses that belong on the inner agent's span.

Two options:

• narrow it to the human's actual answers (confirmed, user_input, selected_choices, edited_output, rejection_feedback) with no framework metadata, or
• park it until open-telemetry/semantic-conventions-genai#188 settles how workflow steps and HITL should look.