GateFox 1.8.8

• Chat Completions servers (local Qwen through vLLM) now report every context overflow
  as context_length_exceeded, including vLLM's "max_tokens must be at least 1, got 0"
  for a prompt that fills the window exactly and overflow wording without the usual token
  counts. Codex then compacts instead of ending the turn with an upstream error.
• A conversation that keeps overflowing a Chat Completions provider stops reaching it
  after 25 consecutive refusals within ten minutes. The client gets one non-retryable
  error that asks for a fresh worker with a compact handoff, instead of trimming one
  history item per request for an hour. Any completed request resets the count.