[Bug] Anthropic API Error: False positive reasoning_extraction safeguard on Opus 5.5
Bug Description
False positive from the reasoning_extraction safeguard on Opus 5.5.
Task: design project in Claude Code. I asked Claude to create process.md, a design decision log for my project (what we decided, why, what options we rejected). I need it to explain decisions to my team and later write a case study.
Opus blocked this request several times, even after I rephrased it. Sonnet did it without problems. After I switched back to Opus, the same session kept failing randomly, also on normal tasks. I think the session context (and process.md itself) now triggers the filter.
Request IDs:
req_011CfpVCMJWcwFY35g4Nsov9
req_011CfpVKJVHfZyanj7Xb3VyP
req_011CfpW9JWcTXRHRjUKTsNQm
I want to keep using Opus for this project. Please check this filter for normal design work.
Environment Info
- Platform: darwin
- Terminal: Apple_Terminal
- Version: 2.1.281
- Feedback ID: 11b4b41a-bec6-4fdd-8fc2-88a259ef77a9
Errors
[]