False positive reasoning_extraction safeguard block on documentation feedback request

Bug Description

Hello,

I'm reporting some false-positive safeguard blocks so you have the context, in case they're ever reviewed against my account.

On 9 October 2026 I was testing my own software, an open web-portal platform I develop. In a local development environment I asked a fresh Claude Code session to build two small page components through my product's MCP server, and to write down which documentation it had been missing, so I could improve our docs.

The request was blocked on Opus 5.5 and then on Sonnet 5.5, with the detail [reasoning_extraction]. A follow-up message with the same documentation-feedback request was blocked as well. The wording was probably the trigger: I asked the agent to note "where it hesitated or guessed" and "why it chose" its approach. I meant usability feedback on my documentation, not the model's internal reasoning.

I stopped as soon as I understood the cause, and I won't resend that wording. Future tests will ask only for observable facts (which tool calls failed, which information was missing).

Request ID: req_011CfrW4rFCE9C9HTH86Fhbm
Message ID: msg_011CfrW4rWpN1VSbEQcjhTCs

Thank you.

Environment Info

• Platform: linux
• Terminal: gnome-terminal
• Version: 2.1.295
• Feedback ID: 2ca33dc0-aa4d-41c8-a5ed-cd50b23462ba