Forced submit tool_choice is rejected by Claude Opus 5.5 and Fable 5.2 (HTTP 400)
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 25/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Stale
- Tech stack
- typescript
- Domain
- ai-infra-agents
Research direction
Read src/llm/pi-runner.ts around line 1962 and the provider model lookup in src/provider/provider-services.ts around lines 491–506. Trace how tool_choice is selected for Anthropic requests, then inspect existing tests for request construction and error handling. Done means the two named models send an accepted choice by default, existing providers and older models retain forced-submit behavior, and validation, retry, and provenance remain unchanged.
Written by the indexing model from the issue text.
Description
Problem
On the direct Anthropic provider, codegenie's default structured-submit calls force the submit tool (tool_choice of type tool or any) while adaptive thinking is on. Claude Opus 5.5 and Fable 5.2 reject that request:
HTTP 400 invalid_request_error: tool_choice: type "tool" and "any" are not supported for this model
so every review with these models fails. The default setup path picks one of them without the user typing its ID: with only Anthropic configured, codegenie provider use opus resolves to anthropic/claude-opus-5-5, and the post-login default goes through the same lookup (src/provider/provider-services.ts:491-506).
The forced choice is set at src/llm/pi-runner.ts:1962. Pi 1.5.2 passes it through unchanged and has no fallback for this error. The earlier "Anthropic falls back" behaviour (PR #38) concerns strict schema sampling, which is a different field.
Evidence
Live check on 2026-11-19, against Anthropic's token-count endpoint only (no generation calls), using the request shape codegenie 0.8.0 actually sends (the stage-5 submit_plan tool):
| model | tool_choice | thinking | result |
|---|---|---|---|
| claude-opus-5-5 | forced | adaptive | 400, as above |
| claude-opus-5-5 | auto | adaptive | 200 |
| claude-fable-5-2 | forced | adaptive | 400, as above |
| claude-fable-5-2 | auto | adaptive | 200 |
| claude-opus-5 | forced | disabled | 200 (the model CI uses, which is why CI passes) |
Vendor documentation: https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 and https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2
Workaround for users today: [llm] forceSubmitToolChoice = false.
Acceptance criteria
- With default settings, reviews on
anthropic/claude-opus-5-5andanthropic/claude-fable-5-2send a tool_choice the model accepts. - Older Claude models (for example Opus 5) and other providers keep their current forced-submit behaviour.
- Submit validation, retry and provenance behaviour are unchanged.
- A test reproduces the 400 request shape for the new models and fails without the fix.
- Before merge, one live end-to-end review on a tiny diff with each of the two models succeeds (owner-approved cost).
Found during the #44 deep review.
- Dominant language
- TypeScript
- Stars
- 8
- Forks
- 1
- Avg merge
- 5m
- Merged PRs (30d)
- 2
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from 0xPolygon/codegenie
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
All issues in 0xPolygon/codegenie
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
Maintainers usually reply within 1 day
-
feature request needs triage
Difficulty 1/5 1-3 hours Newbie friendliness 69/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
HarperFast/symphony#91 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
solidjs/solid-docs#1716 ·
Maintainers usually reply within 1 day