Repository navigation
agent-eval: the nightly bills the subscription only - #437
Merged
Merged
Conversation
The schedule now runs the reference replay and the Claude Code driver (CLAUDE_CODE_OAUTH_TOKEN). The embedded agent calls the API on ANTHROPIC_API_KEY and was metering a Sonnet pass over the 28 tasks every night; it is scored only on a dispatched run with the new `embedded` input ticked. A subscription token is honoured for Claude Code alone, so the embedded agent cannot be moved onto it: a subscription-funded score of the agent is the claude-code job. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CfvCiPHA9NSzKtisp2RGDm
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
The nightly
agent-evalschedule now runssolutionsandclaude-codeonly. Themodeljob (the embedded agent,anthropic.Anthropic()on theANTHROPIC_API_KEYsecret) runs only on aworkflow_dispatchwith the newembeddedboolean input ticked, and still skips with a notice when the secret is missing.Why
The embedded agent was metering a full Sonnet pass over the 28 tasks every night (~230K uncached input + 1.7M cache-read tokens per run). It cannot be moved onto the subscription: a
claude setup-tokenOAuth token is honoured for Claude Code itself, not for the Messages API through the SDK, so a subscription-funded score of the agent is by definition theclaude-codejob, which already runs nightly onCLAUDE_CODE_OAUTH_TOKEN.Docs
README status row,
docs/integrations.md,docs/roadmap.mdsection 5 and the changelog say the same thing. No code or test changes;tests/test_agent_eval.pypasses locally.🤖 Generated with Claude Code
https://claude.ai/code/session_01CfvCiPHA9NSzKtisp2RGDm