:bug: Bug Report: Gemini models (Flash/Pro) stalling with "Historical context" leaks
I'm seeing a recurring issue when using Google Gemini models (via antigravity or gemini-cli). After a few exchanges, the model stops executing tools and instead outputs a line like this at the end of its message:
"[Historical context: a different model called tool "X" with arguments: { ... }. Do not mimic this format - use proper function calling.]"
When this happens, the agent stalls completely. It doesn't actually call the tool, and I have to send a manual message (like "continue" or "halo") to wake it up and make it try again.
Details:
- Happens on both Gemini 3 Flash and Pro.
- It seems the model is "leaking" internal context/instructions instead of emitting a valid tool call.
- The prompt includes a warning "Do not mimic this format", but the model does exactly that.
Is this a known issue with the system prompt or how historical context is injected for Gemini? How can I prevent the model from getting stuck in this "hallucination" loop?
Version: 2026.2.9
"[Historical context: a different model called tool "X" with arguments: { ... }. Do not mimic this format - use proper function calling.]"
When this happens, the agent stalls completely. It doesn't actually call the tool, and I have to send a manual message (like "continue" or "halo") to wake it up and make it try again.
Details:
- Happens on both Gemini 3 Flash and Pro.
- It seems the model is "leaking" internal context/instructions instead of emitting a valid tool call.
- The prompt includes a warning "Do not mimic this format", but the model does exactly that.
Is this a known issue with the system prompt or how historical context is injected for Gemini? How can I prevent the model from getting stuck in this "hallucination" loop?
Version: 2026.2.9
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Similar Threads
Was this page helpful?
