Guardrail Fallback Linter

What it does

With Claude Fable 5, a call in a high-risk domain can be blocked and fall back to Claude Opus 4.8. Code that parses the response blind will break, or worse, quietly run on a downgraded answer nobody decided to accept. Guardrail Fallback Linter finds the call sites where that can happen.

It scans your source for LLM calls that assume every response is a usable answer: a call with no error or refusal branch, a response parsed with no guard, a hardcoded model id with no fallback handling. You get each finding so you can handle the block and the fallback on purpose instead of discovering it in production.

It reads files only. It does not run your code, call any model, or touch the network.

What it checks

• Calls with no error or refusal branch.
• Responses used or parsed with no guard.
• Hardcoded model ids with no fallback handling.
• Python and JavaScript or TypeScript call sites.
• An editable list of client call patterns you can extend.
• Each finding with a rule id, severity, file, line, message, and the evidence line.

What the output looks like

One thing this is not

This is a reliability check. It helps you handle a guardrail block and a model fallback gracefully so your app degrades on purpose. It is not, and must not be used as, a way to bypass, weaken, or evade any safety guardrail.

Questions