Preflight Checklist
What's Wrong?
Hello. I am experiencing what is either a highly misleading UI or a bypass of Fable 5's guardrails.
I have auto-switch to opus 4.8 on false in config.
When I am coding with Fable 5, I tell it to make a sub-agent go implement our code changes. This sub-agent is spawned using Fable 5. The main agent waits, idle. While executing, the sub-agent will suddenly report a guardrail violation and ask me if I want to switch to opus 4.8 or change the prompt and retry. This usually happens on the very first message to the sub-agent but can be mid-implementation. I click change prompt and retry. The sub-agent just continues working, no different prompt, no interjection from me at all. Still on Fable 5.
So either one of three things is happening:
- I am getting false guardrail activation when they really didn't activate at all.
- The guardrail is activating, it switches to Opus 4.8 without my permission, while the UI reports Fable 5.
- This is a legitimate bypass of the Fable 5 guardrail.
What Should Happen?
The sub-agent should pause and allow me to change the prompt. If the same prompt is sent, it should trigger the guardrails again.
Error Messages/Logs
Steps to Reproduce
I pretty much explained in what is wrong.
Claude Model
Other
Is this a regression?
No, this never worked
Last Working Version
No response
Claude Code Version
2.1.1.99
Platform
Anthropic API
Operating System
Ubuntu/Debian Linux
Terminal/Shell
Other
Additional Information
No response
Preflight Checklist
What's Wrong?
Hello. I am experiencing what is either a highly misleading UI or a bypass of Fable 5's guardrails.
I have auto-switch to opus 4.8 on false in config.
When I am coding with Fable 5, I tell it to make a sub-agent go implement our code changes. This sub-agent is spawned using Fable 5. The main agent waits, idle. While executing, the sub-agent will suddenly report a guardrail violation and ask me if I want to switch to opus 4.8 or change the prompt and retry. This usually happens on the very first message to the sub-agent but can be mid-implementation. I click change prompt and retry. The sub-agent just continues working, no different prompt, no interjection from me at all. Still on Fable 5.
So either one of three things is happening:
What Should Happen?
The sub-agent should pause and allow me to change the prompt. If the same prompt is sent, it should trigger the guardrails again.
Error Messages/Logs
Steps to Reproduce
I pretty much explained in what is wrong.
Claude Model
Other
Is this a regression?
No, this never worked
Last Working Version
No response
Claude Code Version
2.1.1.99
Platform
Anthropic API
Operating System
Ubuntu/Debian Linux
Terminal/Shell
Other
Additional Information
No response