Tool-call rewrite
Check if an LLM proxy rewrites tool calls
See whether package-install commands are altered before they reach the client.
How this check works
A tool-call rewrite check looks at whether an LLM proxy changes package-install commands on the way back to you. That change can turn a trusted install into a supply-chain risk.
01Probe
Ask the model for a package-install command through a tool call.
02Inspect
Compare the returned command with the expected install target.
03Conclude
Flag a rewrite when the package name or source changed on the way back.
- How to read the result
- A rewrite is a proxy-layer change, not a model typo. Even a one-character package swap is a supply-chain finding.
- When to run this check
- Use this if your app executes tool calls from the API. Chat-only endpoints still benefit, but agent stacks need it more.
Recent reports
Public reports that included this check.
- sub.callai.onegpt-5.6-solNormalReport timeAug 18
- sub.callai.onegpt-5.6-solBlockedReport timeAug 18
- ai.databyte.co.iddatabyte-m1NormalReport timeAug 18
- ai.databyte.co.iddeepseek-v4-flashNormalReport timeAug 18
- ai.databyte.co.idMiniMax-M3InconclusiveReport timeAug 18
- tokengate-cqt9ivzs.manus.spaceclaude-opus-5NormalReport timeAug 17
FAQ
What is an LLM proxy tool-call rewrite?
It is when the return path changes a tool argument, such as a package-install command, before the client receives it.
Why are package-install commands the test case?
They are a high-impact example. A small rewrite can install a different package and create a supply-chain problem.
Does this check every tool the model can call?
No. It checks this rewrite surface. Other tool arguments can still be changed and need their own review.