Function _warn_if_truncated

openkb/agent/compiler.py:496–515

Warn when the LLM hit the max_tokens cap; return True if it did. ``json_repair`` will silently salvage the truncated prefix, so without this the caller can't tell a short response from a cut-off one. Callers that write a page from the response can pass ``raise_on_truncation=True`` t

(response, step_name: str, max_tokens: int | None)

def _warn_if_truncated(response, step_name: str, max_tokens: int | None) -> bool:
    """Warn when the LLM hit the max_tokens cap; return True if it did.

    ``json_repair`` will silently salvage the truncated prefix, so without
    this the caller can't tell a short response from a cut-off one. Callers
    that write a page from the response can pass ``raise_on_truncation=True``
    to ``_llm_call``/`_llm_call_async`` to turn a truncated response into a
    skip instead of persisting partial content.
    """
    try:
        finish_reason = response.choices[0].finish_reason
    except (AttributeError, IndexError):
        return False
    if finish_reason != "length":
        return False
    cap = f" (max_tokens={max_tokens})" if max_tokens else ""
    logger.warning("LLM [%s] hit length limit%s — output may be truncated.", step_name, cap)
    sys.stdout.write(f"    [WARN] {step_name} hit length limit{cap} — output may be truncated.\n")
    sys.stdout.flush()
    return True


def _parse_json(text: str) -> list | dict: