Cursor rule Cursor
When accuracy improvements may trade against task-inference speed.
15 2mo ago A 305 tokens
blackbox-e2e
02Cursor rule Cursor
E2E only — one Playwright spec, no unit tests.
15 2mo ago A 1,879 tokens
Cursor rule Cursor
Capture pipeline must stay generic — no fixture or ShopDemo tuning.
15 2mo ago A 284 tokens
Cursor rule Cursor
Everything runs in the browser — no server-side inference or app logic.
15 2mo ago A 666 tokens
Cursor rule Cursor
Green Circle commit message format for perf wins.
15 2mo ago A 208 tokens
Cursor rule Cursor
Voice transcripts and the Goal "Run task" input go through one LLM path: the ShowUI-2B UI Navigation mode from the model card (src/actions/navigation.js) — task + screenshot → {'action', 'value', 'position'} dicts.
15 2mo ago A 245 tokens
Cursor rule Cursor
Mind2Web offline grounding eval — strict bbox, one query, no eval oracle.
15 2mo ago A 0 tokens
Cursor rule Cursor
Strictly forbid hardcoding fixture-specific knowledge (selectors, IDs, labels, page layout) in core source files (e.g., src/browser-tools/catalog.js, src/browser-tools/dom-actions.js).
15 2mo ago A 264 tokens
Cursor rule Cursor
Client-side grounding from screenshot only — never live DOM.
15 2mo ago A 683 tokens
Cursor rule Cursor
No query rewrites, candidate sweeps, or coord scoring for navigation.
15 2mo ago A 540 tokens
Cursor rule Cursor
Strictly avoid modifying the source document's CSS or injecting any styles (including tags) into the capture target or its clones.
15 2mo ago A 215 tokens
Cursor rule Cursor
Tool execution acts at the model's grounded point — no label-text DOM lookup.
15 2mo ago A 383 tokens
vla-registry
13Cursor rule Cursor
Multi-model GUI/VL registry — browser wllama grounding (ShowUI-2B is default + E2E gate).
15 2mo ago A 468 tokens
wllama-only
14Cursor rule Cursor
VLA models via wllama in the browser only — never Node/Python inference.
15 2mo ago A 537 tokens
Skill Claude CodeCodexCursor
Run the MiniWoB eval and iterate on raising the success rate without increasing task-inference latency. Use when the user asks to improve, measure, or iterate on MiniWoB (miniwob/miowob) scores or browser tool-interaction eval results.
15 2mo ago A 60 tokens
Instructions file Gemini CLI
Gemini CLI instructions for pdufour/browser-use-wasm, covering project instructions, rule imports, capture: research-first mandate and capture: no visual-diff overfitting.
15 2mo ago A 738 tokens