## What happened?

When a model is configured with `supports_system_message: False` (e.g. the `chatgpt/` provider, whose backend rejects system messages with `{"detail":"System messages are not allowed"}`), LiteLLM maps the system message into the next user message via `map_system_message_pt`.

That function merges contents with plain string concatenation:

```python next_m["content"] = m["content"] + " " + next_m["content"] ```

If the system or user content is a **list of content blocks** — the format sent by Anthropic-style clients (e.g. Claude Code / Claude Agent SDK talking to the proxy's `/v1/messages` endpoint) — this raises:

``` litellm.APIConnectionError: ChatgptException - can only concatenate list (not "str") to list ```

## Minimal reproduction (no network needed)

```python from litellm.litellm_core_utils.prompt_templates.factory import map_system_message_pt

map_system_message_pt(messages=[ {"role": "system", "content": [{"type": "text", "text": "Be brief."}]}, {"role": "user", "content": [{"type": "text", "text": "Hi"}]}, ]) # TypeError: can only concatenate list (not "str") to list ```

## Real-world scenario

Claude Code / Claude Agent SDK → LiteLLM proxy …