feat(plugins): Vision Pack v0 — full-screen screenshot for AI fallback - #28
Merged
Conversation
rrader26
force-pushed
the
feat/vision-screenshot
branch
from
May 12, 2026 12:34
89457aa to
e7b8f22
Compare
Layer 2 of the desktop driver stack: when the accessibility tree is too
sparse to drive an app reliably (custom-drawn UIs, weak-a11y Electron
apps, games), the agent grabs a screenshot and the multimodal AI parses
the image directly.
Tools shipped (1):
agentmark_screenshot full-screen PNG; inline base64 or write-to-path
Capture utilities per platform (no external Node deps):
macOS: screencapture -x (always installed)
Windows: PowerShell + .NET System.Drawing (built-in to .NET 8)
Linux: gnome-screenshot OR scrot (install one)
Returned shape:
{ image_base64?, path?, bytes, captured_by, platform }
Inline base64 by default; pass `output_path` to write to disk and get
back just { path, bytes } — useful for large captures where the agent
doesn't need the raw bytes in-band.
Out of scope (planned follow-up):
- Per-window capture (needs bridge work: UIA / AXAPI window handle
→ PNG buffer)
- OmniParser-style structured GUI parsing inside the plugin
- Auto-fallback on agentmark_desktop_snapshot when the tree is sparse
Tests (3 new, 345 total on this branch): plugin registration. The
screenshot itself needs a real GUI session, so it's smoke-tested via
the live demo paths rather than unit tests.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
rrader26
force-pushed
the
feat/vision-screenshot
branch
from
May 12, 2026 12:35
e7b8f22 to
32bbe90
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Why
Closes the accessibility-coverage gap from the strategic discussion: 99% of well-designed apps expose a usable a11y tree, but custom-drawn UIs, weak-a11y Electron apps, and games don't. Screenshot + multimodal vision = universal fallback. No per-app engineering needed — Claude's own vision handles parsing.
What's in v0
What's NOT in v0 (deliberate)
Usage
```ts
import { createMcpServer, createDesktopPlugin, createVisionPlugin } from '@thinkfleet/agentmark'
const vision = createVisionPlugin()
createMcpServer({ plugins: [desktop, vision, meta] })
```
When the agent gets an empty / sparse snapshot, it can fall back:
```
agentmark_screenshot { output_path: "/tmp/fallback.png" }
→ multimodal AI vision-parses the image with its native vision model
```
Test plan
🤖 Generated with Claude Code