If you attach an image as context for a background agent it appears to work, but when the agent starts doing it’s work it says “I don’t see any images in our current conversation.”
Background agents should be able to receive the same context as normal agents.
This might be specific to the chat-upload path. A tool call that returns image content in its result could go through Cursor differently than a manually attached image, worth testing before assuming it’s a hard wall. Built an MCP server that hands screenshots to agents as structured data, not sure yet if that clears this exact background-agent gap though: https://slimsnap.ai/connector