Skip to content

feat(anthropic): Record finish reasons in AI monitoring spans#5678

Open
ericapisani wants to merge 2 commits intomasterfrom
py-2136-take-2
Open

feat(anthropic): Record finish reasons in AI monitoring spans#5678
ericapisani wants to merge 2 commits intomasterfrom
py-2136-take-2

Conversation

@ericapisani
Copy link
Member

Add GEN_AI_RESPONSE_FINISH_REASONS span data to the Anthropic integration by capturing stop_reason from API responses.

For non-streaming responses, the stop_reason is read directly from the Message result. For streaming responses, it's extracted from the MessageDeltaEvent delta and passed through the _collect_ai_data helper.

This brings the Anthropic integration in line with the OpenAI integration's finish reason tracking.

Capture the stop_reason from Anthropic API responses and set it as
GEN_AI_RESPONSE_FINISH_REASONS span data. Works for both streaming
(via MessageDeltaEvent) and non-streaming responses.

Co-Authored-By: Claude <noreply@anthropic.com>
@linear-code
Copy link

linear-code bot commented Mar 16, 2026

@github-actions
Copy link
Contributor

github-actions bot commented Mar 16, 2026

Semver Impact of This PR

🟡 Minor (new features)

📋 Changelog Preview

This is how your changes will appear in the changelog.
Entries from this PR are highlighted with a left border (blockquote style).


New Features ✨

Anthropic

  • Record finish reasons in AI monitoring spans by ericapisani in #5678
  • Emit gen_ai.chat spans for asynchronous messages.stream() by alexander-alderman-webb in #5572
  • Emit AI Client Spans for synchronous messages.stream() by alexander-alderman-webb in #5565
  • Set gen_ai.response.id span attribute by ericapisani in #5662
  • Add gen_ai.system attribute to spans by ericapisani in #5661

Pydantic Ai

  • Support ImageUrl content type in span instrumentation by ericapisani in #5629
  • Add tool description to execute_tool spans by ericapisani in #5596

Other

  • (crons) Add owner field to MonitorConfig by julwhitney13 in #5610
  • (otlp) Add collector_url option to OTLPIntegration by sl0thentr0py in #5603

Bug Fixes 🐛

  • (ai) Truncate list-based message content in AI monitoring by ericapisani in #5631
  • (anthropic) Close span on GeneratorExit by alexander-alderman-webb in #5643
  • (celery) Propagate user-set headers by sentrivana in #5581
  • (langchain) Wrap finish_reason in array for gen_ai span attribute by ericapisani in #5666
  • (profiler) Prevent buffer race condition during rapid start/stop cycles by ericapisani in #5622
  • (utils) Avoid double serialization of strings in safe_serialize by ericapisani in #5587
  • Enable unused import ruff check and fix unused imports by sentrivana in #5652

Documentation 📚

  • (openai-agents) Remove inapplicable comment by alexander-alderman-webb in #5495
  • Add AGENTS.md by sentrivana in #5579
  • Add set_attribute example to changelog by sentrivana in #5578

Internal Changes 🔧

Anthropic

  • Check system and response ID attributes on spans created by stream() by alexander-alderman-webb in #5665
  • Skip accumulation logic for unexpected types in streamed response by alexander-alderman-webb in #5564
  • Factor out streamed result handling by alexander-alderman-webb in #5563
  • Stream valid JSON by alexander-alderman-webb in #5641
  • Stop mocking response iterator by alexander-alderman-webb in #5573

Docs

  • Remove agentic codebase documentation workflows by dingsdax in #5655
  • Switch agentic workflows from Copilot to Claude engine by dingsdax in #5654
  • Add agentic workflows for codebase documentation by dingsdax in #5649

Openai Agents

  • Do not fail on new tool fields by alexander-alderman-webb in #5625
  • Stop expecting a specific function name by alexander-alderman-webb in #5623
  • Set streaming header when library uses with_streaming_response() by alexander-alderman-webb in #5583
  • Replace mocks with httpx for streamed responses by alexander-alderman-webb in #5580
  • Replace mocks with httpx in non-MCP tool tests by alexander-alderman-webb in #5602
  • Replace mocks with httpx in MCP tool tests by alexander-alderman-webb in #5605
  • Replace mocks with httpx in handoff tests by alexander-alderman-webb in #5604
  • Replace mocks with httpx in API error test by alexander-alderman-webb in #5601
  • Replace mocks with httpx in non-error single-response tests by alexander-alderman-webb in #5600
  • Remove test for unreachable state by alexander-alderman-webb in #5584
  • Expect namespace tool field for new openai versions by alexander-alderman-webb in #5599

Other

  • (graphene) Simplify span creation by sentrivana in #5648
  • (httpx) Resolve type checking failures by alexander-alderman-webb in #5626
  • (pyramid) Support alpha suffixes in version parsing by alexander-alderman-webb in #5618
  • (rust) Don't implement separate scope management by sentrivana in #5639
  • (strawberry) Simplify span creation by sentrivana in #5647
  • 🤖 Update test matrix with new releases (03/16) by github-actions in #5671
  • Remove custom warden action by sentrivana in #5653
  • Add httpx to linting requirements by alexander-alderman-webb in #5644
  • Remove CodeQL action by sentrivana in #5616
  • Normalize dots in package names in populate_tox.py by alexander-alderman-webb in #5574
  • Do not run actions on potel-base by sentrivana in #5614

🤖 This preview updates automatically when you update the PR.

@github-actions
Copy link
Contributor

github-actions bot commented Mar 16, 2026

Codecov Results 📊

13 passed | Total: 13 | Pass Rate: 100% | Execution Time: 8.34s

All tests are passing successfully.

❌ Patch coverage is 0.00%. Project has 14283 uncovered lines.

Files with missing lines (1)
File Patch % Lines
anthropic.py 6.48% ⚠️ 361 Missing

Generated by Codecov Action

@ericapisani ericapisani marked this pull request as ready for review March 16, 2026 14:51
@ericapisani ericapisani requested a review from a team as a code owner March 16, 2026 14:51
Copy link

@cursor cursor bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Fix All in Cursor

Bugbot Autofix prepared a fix for the issue found in the latest run.

  • ✅ Fixed: Unrelated default_integrations=False added to one test only
    • Removed the accidentally committed default_integrations=False parameter from test_streaming_create_message_async to match all other streaming tests.

Create PR

Or push these changes by commenting:

@cursor push f7619664d2
Preview (f7619664d2)
diff --git a/tests/integrations/anthropic/test_anthropic.py b/tests/integrations/anthropic/test_anthropic.py
--- a/tests/integrations/anthropic/test_anthropic.py
+++ b/tests/integrations/anthropic/test_anthropic.py
@@ -508,7 +508,6 @@
     sentry_init(
         integrations=[AnthropicIntegration(include_prompts=include_prompts)],
         traces_sample_rate=1.0,
-        default_integrations=False,
         send_default_pii=send_default_pii,
     )
     events = capture_events()

This Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.

if response_id is not None:
span.set_data(SPANDATA.GEN_AI_RESPONSE_ID, response_id)
if finish_reasons is not None:
span.set_data(SPANDATA.GEN_AI_RESPONSE_FINISH_REASONS, finish_reasons)
Copy link
Contributor

@alexander-alderman-webb alexander-alderman-webb Mar 16, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since anthropic always generates a single output (i.e., no candidates) we can work with the single finish reason in _collect_ai_data (i.e., keep track of a string rather than a list).

If we only convert the finish reason to list form in _set_output_data the preceding code is easier to reason about.

Suggested change
span.set_data(SPANDATA.GEN_AI_RESPONSE_FINISH_REASONS, finish_reasons)
finish_reason: "str | None" = None,
) -> None:
"""
Set output data for the span based on the AI response."""
span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, model)
if response_id is not None:
span.set_data(SPANDATA.GEN_AI_RESPONSE_ID, response_id)
if finish_reason is not None:
span.set_data(SPANDATA.GEN_AI_RESPONSE_FINISH_REASONS, [finish_reason])

Comment on lines +226 to +228
stop_reason = getattr(event.delta, "stop_reason", None)
if stop_reason is not None:
finish_reasons = [stop_reason]
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

non-blocking nitpick: We don't need the getattr(), since event.type == "message_delta" forces event to be an instance of MessageDeltaEvent.

Suggested change
stop_reason = getattr(event.delta, "stop_reason", None)
if stop_reason is not None:
finish_reasons = [stop_reason]
if event.delta.stop_reason is not None:
finish_reason = event.delta.stop_reason

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants