<think> tag will lead to "IndexError: list index out of range"

[Stark-X]

i've tried to use the "minimaxai/minimax-m2.1" from nvidia api, it will return content with <think> in " full_completion" field, then the ouput break with error: IndexError: list index out of range

[stone-w4tch3r]

Hello! I encountered a bug in shellgpt, fixed it locally and asked my claude code to report it and open PR on my behalf

----text below is written by AI, but checked and approved by me, same as the fix and repro steps----

I'm hitting the same IndexError: list index out of range at handler.py:117. In my case the root cause is not <think> tags — it's that API proxies send a final streaming chunk with choices=[] containing only usage metadata.

Reproduced with:

• OpenRouter (openrouter.ai/api/v1) — openrouter/free model
• Kilo Code Gateway (api.kilo.ai/api/gateway) — kilo-auto/small model

The crashing chunk looks like:

ChatCompletionChunk(
    choices=[],  # <-- empty, no delta
    usage=CompletionUsage(completion_tokens=2, prompt_tokens=98, total_tokens=100),
)

This is valid per the OpenAI streaming spec when stream_options.include_usage is enabled, which these proxies do by default.

The fix already exists in main — PR #738 (commit 880e7db) adds the exact guard needed:

if not chunk.choices:
    continue

Could a new release be cut to include this fix? The latest release (1.4.5) predates the fix, so everyone using OpenRouter, Kilo, or similar proxies hits this crash.

Previously reported in #457 (closed as "fix your backend"), but this affects major API providers, not custom backends.

Docker reproduction (if needed)

FROM python:3.13-slim
RUN pip install --no-cache-dir shell-gpt==1.4.5
CMD ["sgpt", "say hello in one word"]

# .env
OPENAI_API_KEY=<your-openrouter-key>
API_BASE_URL=https://openrouter.ai/api/v1
DEFAULT_MODEL=openrouter/free

docker build -t sgpt-bug .
docker run --env-file .env sgpt-bug
# Prints response, then crashes with IndexError