TypeError: expected string or bytes-like object, got 'list' when sending a message with an image attachment

[chhotu-claw]

Summary

Sending a chat message that includes an image attachment (e.g. a screenshot) crashes the chat task with TypeError: expected string or bytes-like object, got 'list'. Text-only messages work fine. The failure happens during $skill-name mention parsing, which assumes the user message content is always a string.

Environment

• cptr 0.6.1
• Python 3.11
• Skills enabled in the workspace (required to hit the code path)

Steps to reproduce

1. Have at least one skill available in the workspace.
2. Open a chat and attach an image (screenshot) to a message.
3. Send the message.
4. The task errors out immediately; the assistant never responds.

Expected

The message (text + image) is processed normally, the same as a text-only message.

Actual

The chat task raises:

File ".../cptr/utils/chat_task.py", line 1177, in run_chat_task
    mentioned = _re.findall(
File ".../re/__init__.py", line 216, in findall
TypeError: expected string or bytes-like object, got 'list'

The offending content value at the time of failure:

[{'type': 'text', 'text': 'getting this error'},
 {'type': 'image', 'media_type': 'image/png', 'base64': 'iVBORw0KGgo...'}]

Root cause

In run_chat_task (cptr/utils/chat_task.py), the $skill-name mention parser runs re.findall directly on last_user["content"]:

mentioned = _re.findall(
    r"\$([a-z0-9](?:[a-z0-9-]*[a-z0-9])?)", last_user["content"]
)

For text-only messages content is a str, but for multimodal messages (any attachment) it's a list of content blocks, which re.findall cannot accept.

Suggested fix

Extract the text blocks before matching:

_content = last_user["content"]
if isinstance(_content, list):
    _text = " ".join(
        b.get("text", "")
        for b in _content
        if isinstance(b, dict) and b.get("type") == "text"
    )
else:
    _text = _content or ""

mentioned = _re.findall(r"\$([a-z0-9](?:[a-z0-9-]*[a-z0-9])?)", _text)

This preserves $skill mention detection (now scanning the text portions of multimodal messages) while fixing the crash. Worth auditing other spots in the message path that assume content is a str, in case the same assumption appears elsewhere.

[tjbck]

Should be addressed!