feat(m3): usage_source=oauth_estimated cost plumbing#14
Open
suzuke wants to merge 1 commit into
Open
Conversation
Closes the loop on PR 19's deferred TODO. ClaudeAgentSDKModel now accumulates `ResultMessage.total_cost_usd` across generate() calls; SmolagentsBackend reads the cumulative estimate and surfaces it via backend_metadata; orchestrator copies onto AttemptNode with `usage_source="oauth_estimated"`. Per spec §4.1 enum, "oauth_estimated" is a new fourth UsageSource value alongside "api" / "cli_estimated" / "unavailable". Disambiguates "API-equivalent estimate of what would have been billed" from "actual metered API cost" — important for the smolagents+claude-subscription path where the user's actual bill is a flat subscription fee. Wiring: - `ledger.UsageSource` Literal extended with "oauth_estimated" - `ClaudeAgentSDKModel._cumulative_cost_estimate_usd` accumulates per ResultMessage; `reset_cumulative_cost()` + property accessor - `SmolagentsBackend.generate_edit` resets at start of agent.run() so cost is per-attempt; reads cumulative after; populates `agent_result.backend_metadata['cost_usd']` + ['usage_source'] - `Orchestrator._make_record` reads backend_metadata; if it has usage_source + cost_usd, those override the default record.usage path (backend knows its own billing model) - LiteLLM path unaffected (no cumulative_cost_estimate_usd attribute) Tests: 8 new in `test_oauth_estimated_cost.py`: - "oauth_estimated" in UsageSource enum - Cumulative cost starts/resets at zero - Cost accumulates across generate() calls - Reset clears for new attempt - backend_metadata includes oauth_estimated cost - backend_metadata omits cost when zero - LiteLLM path doesn't set oauth_estimated metadata (back-compat) Stats: 4 files, +96/-3 LOC. Full suite 2783 passed + 1 pre-existing failure unchanged + 4 skipped, 0 regressions. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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.
Summary
Stacked on #13 (M3 PR 19 smolagents+CC bridge). Closes the deferred TODO from PR 19 round 2: surface API-equivalent cost estimate from
claude_agent_sdk.ResultMessageonto AttemptNode with the newusage_source="oauth_estimated"enum value (per spec §4.1).Why a new enum value? Disambiguates "API-equivalent estimate of what would have been billed" from "actual metered API cost". Important for the smolagents+claude-subscription path where the user's actual bill is a flat subscription fee — postmortem shouldn't mislead them into thinking they're double-billed.
What changed
ledger.UsageSourceLiteral extended: now"api" | "cli_estimated" | "oauth_estimated" | "unavailable". Per-value docs distinguish semantics.ClaudeAgentSDKModelaccumulatesResultMessage.total_cost_usdacrossgenerate()calls;reset_cumulative_cost()+ property accessor.SmolagentsBackend.generate_editresets at start ofagent.run()(per-attempt isolation); reads cumulative after; populatesagent_result.backend_metadatawithcost_usd+usage_source="oauth_estimated"when non-zero.Orchestrator._make_recordreuses_extract_backend_metadata(PR 17 helper); backend_metadata's cost/usage_source override the defaultrecord.usage-derived path.LiteLLM path unaffected — guarded by
hasattrchecks.Reviewer trail
Single round, VERIFIED. Math reconciles (PR 19 baseline 2774 + R2 fix 1 + PR 19a 8 = 2783).
Reviewer also confirmed the two PR 19 round-2 non-blocking nits were closed in interim commit
a15b606(typed auth-error classification + docstring future-tense) — landed between PRs as the established pattern.Stats
git diff --stat HEAD~1 HEAD:tests/test_oauth_estimated_cost.pyTest plan
oauth_estimatedvalue in UsageSource Literal🤖 Generated with Claude Code