| name | screen-share-screenshots |
| description | Look at screenshots the user captured with the Screen Share app. Use when the user refers to their screen, to what they are looking at, to a screenshot they just took, or asks you to read, review, debug, or transcribe something they shared visually rather than pasting. Also use when they ask what screenshots they have taken. |
| metadata | {"emoji":"🖥️","vellum":{"display-name":"Screen Share screenshots","category":"productivity","activation-hints":["User asks what is on their screen, or what you can see","User says they took or just took a screenshot","User asks you to read, review, or debug an error, diff, or design they are looking at","User refers to 'this' or 'here' in a way that only makes sense visually","User asks which screenshots exist, or to compare two of them"],"avoid-when":["User pasted or attached an image directly in the conversation","User wants a screenshot of a web page you can fetch or browse yourself","User is asking about the Screen Share plugin's code rather than its captures"]}} |
The Screen Share app lets the user share a screen and save stills of it with a
"Take screenshot" button. Each capture is a PNG in this plugin's workspace data
directory. Nothing puts those files in front of you automatically — read them
when the conversation calls for it.
Steps
- Run
bun run scripts/list-screenshots.ts from this skill's directory to list
what has been captured, newest first. It prints one screenshot per line as
<captured-at> <size> <absolute-path>, and takes an optional count
(default 10), e.g. bun run scripts/list-screenshots.ts 3.
- Read the screenshot you need with
file_read on its absolute path. It
returns PNGs as images, so you see the pixels — do not try to parse the file
yourself, and do not use bash to cat it.
- Answer from what you see. Quote error text, code, or labels verbatim when the
user is asking you to read something, since that is usually the point.
Choosing which screenshot
Default to the newest one: "what's on my screen" almost always means the capture
the user just took. Reach past it only when the user's wording points elsewhere —
"the one before that", "the screenshot from this morning", "compare the last
two". Read several when a comparison is asked for.
If the list comes back empty, say so plainly and tell the user how to capture
one: open the Screen Share app, click Share screen to pick a screen or
window, then Take screenshot. Do not guess at what their screen shows, and do
not substitute a different screenshot tool.
Notes
- The filename encodes when the capture happened and what was shared, e.g.
screenshot-20260801-161500-chrome-vellum-docs.png. It is a useful hint about
which capture the user means, but the <captured-at> column is authoritative.
- Screenshots are scaled down on capture (1920px wide by default), so very small
text may be soft. If something is genuinely unreadable, say which part rather
than guessing, and suggest raising
maxWidth in the plugin's config.json.
- The user chose each of these captures deliberately — the app streams no video.
Treat a screenshot as something they handed you, and don't extrapolate about
the rest of their screen from it.
