Surface output-limit truncation instead of misformat retries

Record the Chat Completions finish_reason on LLMResult and add a
message_loop_result extension that, when a turn stopped at the output
token limit (finish_reason "length"/"max_tokens") without a complete
tool request, retries with a warning naming truncation as the cause
rather than the generic misformat message. The warning counts toward
the existing unusable-response loop limit.

Context Doctor now leaves results already handled by an earlier
extension untouched, so truncated envelopes are not repaired and
dispatched with cut-off arguments.

Fixes #1836
This commit is contained in:
Mohammad Hijjawi 2026-09-25 23:59:51 +01:00 • committed by Alessandro
parent ff2a5b9c9c
commit 2f8a06515e
17 changed files with 166 additions and 3 deletions

View file

@ -25,6 +25,7 @@ class StopUnusableResponseLoop(Extension):
self.agent.read_prompt("fw.msg_empty_response.md"),
self.agent.read_prompt("fw.msg_reasoning_only.md"),
self.agent.read_prompt("fw.msg_thoughts_fallback.md"),
self.agent.read_prompt("fw.msg_output_limit.md"),
}:
return