[Gemini] rejects tool results with JSON nested deeper than ~31 levels (400 INVALID_ARGUMENT)

## Observed symptom [Bug]: Gemini rejects tool results with JSON nested deeper than ~31 levels (400 INVALID_ARGUMENT); the session stays broken ### Bug Description A tool result whose JSON nests deeper than about 31 levels makes every later request in the session fail on Gemini with a bare `HTTP 400 INVALID_ARGUMENT` ("Request contains an invalid argument."). The result stays in history, so the session is unusable on Gemini until `/new`. The error message gives no hint of the cause. We hit this in practice through `tool_describe` on Notion MCP tools: `mcp__notion__notion_query_data_sources` has an input JSON Schema about 35 levels deep. The primary (Codex) accepted it. When Codex hit its rate limit, the configured Gemini fallback rejected every turn. ### Steps to Reproduce Minimal repro against OpenRouter (the same happens with `google/gemini-3.1-flash-lite`, `google/gemini-2.5-flash` and `google/gemini-3.7-flash`): ```python ## Literal error text ``` 400 INVALID_ARGUMENT nested deeper than ~31 levels ``` ## Environment GitHub issue NousResearch/hermes-agent #129979; state=open ## Trigger / context ### Bug Description A tool result whose JSON nests deeper than about 31 levels makes every later request in the session fail on Gemini with a bare `HTTP 400 INVALID_ARGUMENT` ("Request contains an invalid argument."). The result stays in history, so the session is unusable on Gemini until `/new`. The error message gives no hint of the cause. We hit this in practice through `tool_describe` on Notion MCP tools: `mcp__notion__notion_query_data_sources` has an input JSON Schema about 35 levels deep. The primary (Codex) accepted it. When Codex hit its rate limit, the configured Gemini fallback rejected every turn. ### Steps to Reproduce Minimal repro against OpenRouter (the same happens with `google/gemini-3.1-flash-lite`, `google/gemini-2.5-flash` and `google/gemini-3.7-flash`): ```python import json, requests def nest(n): o = {"leaf": 1} for _ in range(n): o = {"k": o} return o msgs = [ {"role": "user", "content": "describe the tool"}, {"role": "assistant", "content": "", "tool_calls": [{"id": "c1", "type": "function", "function": {"name": "tool_describe", "arguments": "{}"}}]}, {"role": "tool", "tool_call_id": "c1", "content": json.dumps(nest(32))}, ] tools = [{"type": "function", "function": {"name": "tool_describe", "parameters": {"type": "object", "properties": {}}}}] r = requests.post("https://openrouter.ai/api/v1/chat/completions", headers={"Authorization": f"Bearer {KEY}"}, json={"model": "google/gemini-3.1-flash-lite", "messag ## Source evidence - https://github.com/NousResearch/hermes-agent/issues/129979 (2026-10-01T04:42:52Z) — GitHub issue #129979 (open) ## Freshness - classification: FRESH_7D - retrieved_at: 2026-10-02T09:40:08.726088Z - scout_id: Ccb65ca7388 ## Scout notes This Problem was published by the KFA Fresh Friction Scout from public sources. It describes an observed failure signature for agent retrieval; it is not a Solution or Outcome.