feat(weave): trace Claude Agent SDK subagents by rgao-coreweave · Pull Request #7666 · wandb/weave

rgao-coreweave · 2026-07-29T20:19:20Z

Summary

Represent Claude Agent SDK Agent and legacy Task calls as nested Python GenAI SubAgent spans.
Route delegated assistant messages, tools, nested agents, and results through their parent_tool_use_id ancestry.
Preserve model, prompt, result, error, description, and tool-call metadata on each subagent span.
Keep parallel subagents off the ambient OTel context while explicitly parenting their child LLM, tool, and subagent spans.

Stack

master
└── #7666 subagent tracing (this PR)
    └── #7665 async-iterable multi-turn tracing

This PR is directly based on master. #7665 is intentionally stacked on this branch.

Before and after

The same 24-span cassette replay was run on master and this PR. mission_control launches researcher, analyst, and reviewer concurrently; reviewer launches a nested fact_checker. Together they run WebSearch, WebFetch, Bash, Python, Read, and Checklist tools before the root agent synthesizes the result.

In each link, select the Trace tree button (the branch icon) above the trace panel. The default flamegraph groups overlapping siblings, which can obscure the parent hierarchy.

Run	Spans	`invoke_agent` spans	Maximum depth	Result
`master` (`fc63e44005`)	24	1	1	All 23 operation spans are direct children of `mission_control`; delegate calls appear as ordinary `Agent` / `Task` tools.
Subagent-only head (`7c2d3b987d`)	24	5	3	Delegate chats and tools are grouped under their agents, including `mission_control → reviewer → fact_checker → Read`.

The resulting topology is:

mission_control
├── researcher
│   ├── WebSearch
│   └── WebFetch
├── analyst
│   ├── Bash
│   └── Python
└── reviewer
    ├── fact_checker
    │   └── Read
    └── Checklist

The parallel-subagent regression covers delegates closing in completion order rather than launch-stack order. The integration starts them with SubAgent.start(set_current=False): explicitly threaded child spans retain the hierarchy above, while unrelated application work remains under the root turn.

Testing

Source-asserted live Weave replay on master and the subagent-only head; queried span counts, parent IDs, and maximum depths match the table above.
Complete Claude Agent SDK cassette suites (OTel and calls-based) plus the conversation SubAgent suite: 46 passed on Python 3.11.
Ruff 0.15.5 check and format verification passed for all changed Python files.

Breaking changes

None.

rgao-coreweave · 2026-07-29T20:19:32Z

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

This stack of pull requests is managed by Graphite. Learn more about stacking.

codecov · 2026-07-30T03:43:33Z

Codecov Report

❌ Patch coverage is 83.80952% with 34 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../integrations/claude_agent_sdk/otel_integration.py	79.39%	15 Missing and 19 partials ⚠️

📢 Thoughts on this report? Let us know!

w-b-hivemind · 2026-07-30T05:07:39Z

HiveMind Sessions

9 sessions · 34s · $0.13

Session	Agent	Duration	Tokens	Cost	Lines
Delegate Query To Researcher Subagent `ef9998ce-7878-4452-abbd-69f4885f30e8`	claude	7s	245	$0.03	+0 -0
Code Word Memory Test `8c3c91e2-e1d9-4af2-8a19-51f1f86c7a68`	claude	4s	11	$0.01	+0 -0
Identifying Square Color and Number `4b9b11a0-b60a-40d4-957e-b091cc5826e3`	claude	2s	19	$0.01	+0 -0
Researcher Subagent Planet Query `3c2f7791-bf4d-4bc9-bcec-71b23e676766`	claude	4s	241	$0.02	+0 -0
Code Word Memory Test `7d0bf574-da29-4a3c-891d-0f50d3858c37`	claude	3s	20	$0.00	+0 -0
Identify Square Color and Number `f98e3dd3-ac85-4a24-b1db-c2a9d6c747a6`	claude	1s	19	$0.00	+0 -0
Delegating Research Query To Subagent `f5e3f591-72e7-441a-92a8-e700d4c9ce5e`	claude	5s	241	$0.03	+0 -0
Code Word Memory Test `aff0f637-4f11-443b-b080-3d6a25d1d7db`	claude	3s	20	$0.01	+0 -0
Identifying Square Color and Number `816fef1d-a71a-46bf-a6d8-4ff6196d7aa8`	claude	1s	19	$0.01	+0 -0
Total		34s	835	$0.13	+0 -0

Screenshots

View all sessions in HiveMind →

Run claude --resume ef9998ce-7878-4452-abbd-69f4885f30e8 to pickup where you left off.

Parallel Agent/Task delegations close in completion order, not launch order. SubAgent.__enter__ attached the span to the ambient OTel context and end() detaches through ContextVar.reset, which assumes LIFO: closing A before B left the context pointing at ended span A, so user code running between streamed messages nested under a dead span. Add SubAgent.start(set_current=False), which starts the span without pushing it onto the ambient stack, and use it from the Claude Agent SDK integration. Children nest either way, because start_llm / start_tool / start_subagent thread an explicit parent context. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

ro31337

Approving as discussed here https://weightsandbiases.slack.com/archives/C0APFJB2H5E/p1785885124302999?thread_ts=1785849300.988039&cid=C0APFJB2H5E

rgao-coreweave · 2026-08-05T20:32:09Z

Approving as discussed here https://weightsandbiases.slack.com/archives/C0APFJB2H5E/p1785885124302999?thread_ts=1785849300.988039&cid=C0APFJB2H5E

Merging in, thanks for the review!

rgao-coreweave force-pushed the rgao/claude-agent-sdk-python-subagent-tracing branch from 421cd52 to c80f42c Compare July 30, 2026 03:33

rgao-coreweave marked this pull request as ready for review July 30, 2026 05:33

rgao-coreweave requested a review from a team as a code owner July 30, 2026 05:33

rgao-coreweave requested a review from a team July 30, 2026 16:49

ro31337 reviewed Jul 30, 2026

View reviewed changes

Comment thread weave/integrations/claude_agent_sdk/otel_integration.py Outdated

rgao-coreweave force-pushed the rgao/claude-agent-sdk-python-subagent-tracing branch from 23b51b9 to 79d2ff0 Compare July 31, 2026 18:51

rgao-coreweave force-pushed the rgao/claude-agent-sdk-python-subagent-tracing branch from 79d2ff0 to 4f17c21 Compare July 31, 2026 23:01

rgao-coreweave force-pushed the rgao/claude-agent-sdk-python-subagent-tracing branch from 4f17c21 to d03a57b Compare July 31, 2026 23:48

rgao-coreweave changed the base branch from rgao/claude-agent-sdk-python-multi-turn to master August 3, 2026 16:50

rgao-coreweave changed the title ~~feat(weave): trace Claude Agent SDK subagents~~ feat(weave): trace Claude Agent SDK multi-turn and subagent flows Aug 3, 2026

rgao-coreweave and others added 2 commits August 3, 2026 12:47

feat(weave): trace Claude Agent SDK subagents

a04a664

rgao-coreweave force-pushed the rgao/claude-agent-sdk-python-subagent-tracing branch from d03a57b to 7c2d3b9 Compare August 3, 2026 19:50

rgao-coreweave changed the title ~~feat(weave): trace Claude Agent SDK multi-turn and subagent flows~~ feat(weave): trace Claude Agent SDK subagents Aug 3, 2026

rgao-coreweave requested a review from ro31337 August 4, 2026 05:43

ro31337 reviewed Aug 4, 2026

View reviewed changes

Comment thread weave/integrations/claude_agent_sdk/otel_integration.py Outdated

Comment thread weave/integrations/claude_agent_sdk/otel_integration.py Outdated

Comment thread weave/integrations/claude_agent_sdk/otel_integration.py

rgao-coreweave added 2 commits August 4, 2026 15:26

fix(weave): apply content policy to subagent spans

115bdb9

fix(weave): trace background subagent lifecycles

5893a74

ro31337 self-requested a review August 4, 2026 23:19

ro31337 approved these changes Aug 4, 2026

View reviewed changes

rgao-coreweave merged commit 49f8c1d into master Aug 5, 2026
108 checks passed

rgao-coreweave deleted the rgao/claude-agent-sdk-python-subagent-tracing branch August 5, 2026 20:32

github-actions Bot locked and limited conversation to collaborators Aug 5, 2026

Navigation Menu

Uh oh!

feat(weave): trace Claude Agent SDK subagents - #7666

feat(weave): trace Claude Agent SDK subagents#7666
rgao-coreweave merged 4 commits into
masterfrom
rgao/claude-agent-sdk-python-subagent-tracing

rgao-coreweave commented Jul 29, 2026 •

edited

Loading

Uh oh!

rgao-coreweave commented Jul 29, 2026

Uh oh!

codecov Bot commented Jul 30, 2026 •

edited

Loading

Uh oh!

w-b-hivemind Bot commented Jul 30, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ro31337 left a comment

Uh oh!

Uh oh!

rgao-coreweave commented Aug 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

rgao-coreweave commented Jul 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Stack

Before and after

Testing

Breaking changes

Uh oh!

rgao-coreweave commented Jul 29, 2026

Uh oh!

codecov Bot commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

w-b-hivemind Bot commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

HiveMind Sessions

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ro31337 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

rgao-coreweave commented Aug 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rgao-coreweave commented Jul 29, 2026 •

edited

Loading

codecov Bot commented Jul 30, 2026 •

edited

Loading

w-b-hivemind Bot commented Jul 30, 2026 •

edited

Loading