forgewright
Standalone UI automation and vision toolkit for Windows. Captures screens, compresses frames with tile-delta hashing, runs visual gate assessments, and drives browsers and desktop apps programmatically.
Features
- DXGI capture with GDI BitBlt fallback
- BMP decode/encode (24-bit BGR, bottom-up and top-down)
- Tile-hash frame compression — FNV-1a per-tile hashing with delta detection
- Perceptual hashing — 64-bit pHash for image similarity
- Visual gate — automated pass/fail assessment against configurable expectations
- Macro compilation —
SkillMacrotoAgentInputFramecompiler - Shared memory spooler — ring buffer for agent input frames
- Photometric stereo — normal map estimation from multi-light captures
- Sprite generation — contour extraction and verification
- Edge detection — Sobel-based edge density and feature extraction
- CDP automation — Chrome DevTools Protocol WebSocket client
- UIA automation — Windows UI Automation accessibility tree driver
- Burst capture — multi-frame capture with GIF encoding
CLI
forgewright spawn <exe> [args...] # Launch a process
forgewright drive <target> # Drive a running app via CDP/UIA
forgewright login # Interactive login prompt
Library Usage
use VisionBridge;
use Frame;
use ;
// Decode a BMP screenshot
let frame = decode_bmp?;
// Compare two frames perceptually
let h1 = perceptual_hash;
let h2 = perceptual_hash;
let distance = hamming_distance;
Platform Notes
These modules require Windows and are gated behind #[cfg(target_os = "windows")]:
dxgi— DXGI Desktop Duplication + GDI fallbackuia— Windows UI Automationwin32— Win32 window enumerationdual_capture— Combined DXGI + window capture
All other modules (vision pipeline, gate, bridge, compression) are cross-platform.
License
MIT OR Apache-2.0