feat(ai): add single message truncation

[shellmayr]

• Realized in this discussion that we don't handle the case of the single large message well at the moment
• Truncate single messages to size limit minus overhead.
  • If the message is not the last message, this will still result in the message being truncated.
  • If it is the last message, it will be retained, but truncated

Contributes to TET-1208

[cursor]

Bug: In-place Mutation of Message Content Causes Side Effects

The function mutates the original message dict instead of creating a copy before modification. This causes unintended side effects where the caller's input data is modified. Similar to normalize_message_roles, this function should create a copy of the message before modifying its content field.

 

[cursor]

Bug: UTF-8 Truncation Mismatch Causes Encoding Errors

The truncation uses content[:available_content_bytes] which slices by character index rather than byte length. For multi-byte UTF-8 characters like emoji or non-ASCII text, this produces incorrect results since available_content_bytes is a byte count but string slicing operates on character positions. This can truncate too many or too few bytes, and may slice mid-character causing encoding errors.

 

[linear]

TET-1208 Python SDK: truncate gen_ai messages

[alexander-alderman-webb]

Looks good in terms of the general approach.

The message mutation also to be addressed; we should not edit the return value of another library and return the edited value to the user.

[alexander-alderman-webb]

Can we instead check if the returned array has one element, and if so, truncate the single message?

Let me know if I am missing some edge case that wouldn't be covered that way.