fix: shrink images to reported provider dimension limit (#45979)

Parse provider-reported image pixel ceilings so many-image Anthropic requests can recover by shrinking Retina screenshots below the stricter limit instead of retrying the same rejected payload.
This commit is contained in:
brooklyn! 2026-06-14 01:07:43 -05:00 committed by GitHub
parent cdf30a7ac6
commit 4b5ba112ad
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 128 additions and 12 deletions

View file

@ -71,6 +71,35 @@ logger = logging.getLogger(__name__)
INTERRUPT_WAITING_FOR_MODEL_PREFIX = "Operation interrupted: waiting for model response ("
def _image_error_max_dimension(error: Exception) -> Optional[int]:
"""Extract a provider-reported image dimension ceiling, if present."""
parts = []
for value in (
error,
getattr(error, "message", None),
getattr(error, "body", None),
):
if value:
try:
parts.append(str(value))
except Exception:
pass
text = " ".join(parts).lower()
if "image" not in text or "dimension" not in text or "max allowed size" not in text:
return None
match = re.search(r"max allowed size(?:\s+for [^:]+)?:\s*(\d{3,5})\s*pixels?", text)
if not match:
return None
try:
max_dimension = int(match.group(1))
except ValueError:
return None
if 512 <= max_dimension <= 8000:
return max_dimension
return None
def _ollama_context_limit_error(agent: Any, request_tokens: int) -> Optional[str]:
"""Return a user-facing error when Ollama is loaded with too little context."""
if not getattr(agent, "tools", None):
@ -2070,7 +2099,11 @@ def run_conversation(
and not _retry.image_shrink_retry_attempted
):
_retry.image_shrink_retry_attempted = True
if agent._try_shrink_image_parts_in_messages(api_messages):
image_max_dimension = _image_error_max_dimension(api_error) or 8000
if agent._try_shrink_image_parts_in_messages(
api_messages,
max_dimension=image_max_dimension,
):
agent._vprint(
f"{agent.log_prefix}📐 Image(s) exceeded provider size limit — "
f"shrank and retrying...",