mirror of
https://github.com/openai/codex.git
synced 2026-11-15 21:50:46 +00:00
## Why The agents overview submitted background task prompts as plain text, without image attachments or their text elements. ## What changed - Enable image pasting in the overview composer and include attachments and text elements in the first task prompt. - Resolve local image paths and send image bytes when using a remote workspace. - Reject image inputs for models declared text-only before starting a thread. - Restore unsent text and attachments after failures, or show image reattachment paths when a newer draft prevents restoration. ## Testing Add tests for attachment rendering and submission, remote image payloads, draft recovery without overwriting newer text, and text-only model rejection. GitOrigin-RevId: 08e3db8bace51ec007f5741f1b2be94658cbaea4