upload-screenshot

name: upload-screenshot description: Upload an existing screenshot file to the Open-Inspect session

upload-screenshot

Use this skill when you already have a screenshot file on disk and need to upload it to the Open-Inspect session. The screenshot may have been captured by any source: Playwright MCP, agent-browser, a manual file, or any other tool.

For the full browser-open + capture + upload workflow, use visual-verification instead.

Key Fact

upload-media is a bash command installed on PATH. Run it with your Bash tool. It is not an MCP tool or a tool binding.

When To Use It

• Upload a screenshot that was already captured (e.g. via Playwright MCP)
• Upload an image file the user points you to
• The user says "upload the screenshot" or "upload this image"

Required Workflow

1. Confirm the screenshot file exists on disk.
2. Run upload-media via Bash with the file path and metadata flags.
3. Report the returned artifactId to the user.

Command

upload-media <file-path> \
  --caption "Description of screenshot" \
  --source-url "https://example.com" \
  [--full-page] \
  [--annotated] \
  [--viewport '{"width":1512,"height":982}']

All flags except the file path are optional. Include whichever metadata you know:

• --caption — what the screenshot shows
• --source-url — the URL that was captured
• --full-page — set if the screenshot is a full-page capture
• --annotated — set if the screenshot has annotations
• --viewport — JSON object with width and height of the viewport used

Supported File Types

.png, .jpg / .jpeg, .webp

Success Criteria

The task is not complete until:

1. upload-media returned a JSON response containing an artifactId.
2. The artifactId is reported to the user.
3. The response states what was uploaded and the source URL (if known).

Example

Uploaded screenshot of the homepage.
Source: https://example.com
Uploaded artifact: abc123def456

Guardrails

• Do not claim the screenshot was uploaded unless upload-media returned an artifact ID.
• If the file does not exist or is not a supported type, report the error instead of retrying silently.
• If the user needs a full browser workflow (open page, set viewport, capture, upload), delegate to visual-verification instead.