hermes - 💡(How to fix) Fix feat(cli): support image input (file path / clipboard paste) in terminal mode [1 pull requests]

Problem

CLI mode has no image support. If you want the agent to look at a screenshot, you have to type out the full file path and explicitly ask it to analyze. Meanwhile Telegram and Discord just... handle it — drop an image and the agent sees it.

There's no equivalent for terminal users.

Proposed solution

Two ideas, in order of effort:

1. Auto-detect image paths in input

If the user's input contains a path ending in .png/.jpg/.jpeg/.webp/.gif, automatically run vision_analyze on it before sending to the agent.

The agent gets both the image and the question without the user needing to say "look at this image."

2. Clipboard paste (stretch goal)

Detect images in the system clipboard (e.g. via xclip -selection clipboard -t image/png or wl-paste) and analyze on paste or a keybinding like Ctrl+V.

A typical workflow would be:

1. grim -g "$(slurp)" - | wl-copy (screenshot to clipboard)
2. Ctrl+V in hermes CLI
3. Agent sees the image

This would make the CLI feel a lot closer to the messaging platform experience.

Why

• Screenshots are the fastest way to show an agent what's on screen
• CLI and chat platforms should have rough feature parity for core workflows
• Right now "analyze this image" requires knowing the exact path, which is annoying