← All posts
5 min read

Agents Get a Manifest, Teammates Get a Picture

A screenshot sent to Claude and a screenshot sent to Slack should not be the same payload. Screentack sends structured context to agents and a clean image to people — automatically, from one capture.

Most capture tools treat "share" as one action. You get an image and a destination, and what happens next is your problem.

But the two things you actually do with a screenshot are not the same thing at all. Sometimes you are handing it to a machine that needs to act on it. Sometimes you are handing it to a person who needs to glance at it. The payload that serves one is actively wrong for the other.

The two audiences

An agent wants structure. Give Claude or Cursor a flat image and it has to work out which pixel you meant. Give it the full screenshot, a crop of each labelled region, and a manifest of where each region sits, and it knows — because you told it.

A person wants a picture. Drop that same manifest into a Slack thread and your teammate sees a wall of file paths, coordinates, and session IDs wrapped around the one thing they wanted to look at. It is not richer context. It is noise with a screenshot buried in it.

What each one gets

Screentack's send targets come in two kinds, and the kind decides the payload.

AI targets — Claude, ChatGPT, Gemini, Copilot, Cursor, Windsurf, Codex — receive the image plus the metadata that makes it machine-readable.

Communication targets — Slack, Discord, Messages, Mail — receive the image only. Never a manifest. A human on the other end gets a clean annotated picture and nothing to scroll past.

That is eleven defaults, and you can add your own custom targets for anything else you paste into regularly.

The distinction is the point. You are not picking a format every time you share something; the tool already knows whether it is talking to a machine or a person.

When you just want the picture

There is also a dedicated image mode: it puts only the picture on the clipboard, with no agent data attached. Paste it into a doc, a ticket, a design review, an email — anywhere the receiving end is a human and the manifest would be clutter.

Why this matters more than it sounds

The annotation you make is genuinely useful to both audiences. A box around the broken button with "misaligned on mobile" written next to it tells your agent what to fix and tells your colleague what you are pointing at.

What differs is everything wrapped around it. Stripping the machine-readable layer for humans is not dumbing it down — it is the difference between a message someone reads and a message someone skims past.

And it removes the small, constant friction of deciding: no exporting a second time, no deleting a manifest out of a Slack message, no accidentally pasting a file path into a customer thread.

For the other half of this loop — where the agent draws its own understanding back onto your screen so you can correct it before it writes code — see bidirectional visual feedback for AI coding agents.

One capture, the right payload for whoever receives it. Download Screentack — a free 7-day trial, then $29 once.

Frequently asked questions

Where can Screentack send a capture directly?

Screentack ships with eleven default send targets in two groups. AI targets: Claude, ChatGPT, Gemini, Copilot, Cursor, Windsurf, and Codex. Communication targets: Slack, Discord, Messages, and Mail. You can also add your own custom targets.

Does my teammate get the same thing my agent gets?

No, and that is deliberate. Communication targets receive the image only — a manifest full of file paths and coordinates is noise in a Slack thread. AI targets receive the image plus the metadata that makes it machine-readable.

How do I send just the picture, with no agent data?

Image mode puts only the picture on the clipboard, so you can paste it into Slack, Mail, a doc, or anywhere else without the manifest coming along.