Skip to content

Conversation

@nsimmons
Copy link

@nsimmons nsimmons commented Feb 8, 2026

Summary

Fix for #65, and alternate solution from #75

@anthropic-ai/claude-agent-sdk 0.1.0 gem has a hard limit of 32k output tokens. This limit means that setting CLAUDE_CODE_MAX_OUTPUT_TOKENS to > 32k has no impact. Upgrading to the latest version solves this issue as that limitation has been removed.

Changes

  • Upgraded @anthropic-ai/claude-agent-sdk 0.1.0 and dependent zod gems to latest versions.
  • Modified src/ai/claude-executor.ts to pass CLAUDE_CODE_MAX_OUTPUT_TOKENS in the way newest version of claude agent SDK expects.

Test Plan

  • Set CLAUDE_CODE_MAX_OUTPUT_TOKENS to 64k
  • Run ./shannon start on a repo that generates a pre-recon output that is larger than 32k
  • Verify deliverable is created without any output token errors

I understand you're not accepting external PRs. I have found this tool very useful and wanted to do what I can to suggest improvements even if they are not accepted. If it helps you resolve an issue faster then I have done my part to contribute back.

@keygraphVarun
Copy link
Collaborator

Thanks for digging into this and for the detailed write-up!

Upgrading the Claude Agent SDK has been a priority for us for a while, but it comes with a number of breaking changes, so we have been taking the time to do it carefully. We have a preliminary version interally and are still testing it - we should be pushing it later this week, which will fix this issue.

@nsimmons
Copy link
Author

That's great news, thanks for the update.

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