fix(claude_agent_sdk): attribute usage to model call spans#572
Closed
Abhijeet Prasad (AbhiPrasad) wants to merge 1 commit into
Closed
fix(claude_agent_sdk): attribute usage to model call spans#572Abhijeet Prasad (AbhiPrasad) wants to merge 1 commit into
Abhijeet Prasad (AbhiPrasad) wants to merge 1 commit into
Conversation
Claude Agent runs may make several orchestrator and delegated model calls, but cumulative result usage was being assigned to one final span. This hid delegated usage and could price tokens using the wrong model. Before: Claude Agent [TASK] ├── root LLM [opus] metrics: none ├── Agent [TOOL] │ └── subagent [TASK] │ └── LLM [haiku] metrics: none └── final LLM [opus] metrics: cumulative usage for every call After: Claude Agent [TASK] ├── root LLM [opus] metrics: root call usage ├── Agent [TOOL] │ └── subagent [TASK] │ └── LLM [haiku] metrics: delegated call usage └── final LLM [opus] metrics: reconciled final-call remainder Each LLM span now keeps its own metadata.model and token/cache metrics, including cache-read, cache-creation, 5m/1h cache TTL, and server-tool usage. Repeated cumulative snapshots are converted to deltas so trace totals do not double-count them. Result usage also provides a safe fallback when assistant usage is unavailable. Result total_cost_usd remains request metadata on the root task rather than a parent-owned estimated cost. This keeps trace totals and per-model cost estimates accurate across root, subagent, concurrent-agent, and older SDK flows. Extend cassette-backed coverage for multi-turn and concurrent-agent requests on both supported Claude Agent SDK versions.
Member
Author
|
actually this approach is problematic as well. Going back to the drawing board. |
Abhijeet Prasad (AbhiPrasad)
deleted the
abhi-fix-claude-agent-per-span-usage
branch
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supercedes #565
resolves #564
resolves https://linear.app/braintrustdata/issue/SDK-52/claude-agent-sdk-handle-result-logs-main-agent-usage-under-counts
AI Summary:
Claude Agent runs may make several orchestrator and delegated model calls, but cumulative result usage was being assigned to one final span. This hid delegated usage and could price tokens using the wrong model.
Before:
Claude Agent [TASK]
├── root LLM [opus] metrics: none
├── Agent [TOOL]
│ └── subagent [TASK]
│ └── LLM [haiku] metrics: none
└── final LLM [opus] metrics: cumulative usage for every call
After:
Claude Agent [TASK]
├── root LLM [opus] metrics: root call usage
├── Agent [TOOL]
│ └── subagent [TASK]
│ └── LLM [haiku] metrics: delegated call usage
└── final LLM [opus] metrics: reconciled final-call remainder
Each LLM span now keeps its own metadata.model and token/cache metrics, which makes calculations for everything correct.