Chrome DevTools Optimizer Skill
Cuts token cost when driving the Chrome DevTools MCP by preferring text snapshots over screenshots and only escalating to vision when necessary.
Usage
Consult the pattern guides in patterns/ and the token-cost reference in
references/ when performing browser automation:
- Snapshot-first: read the accessibility/text snapshot before taking a screenshot.
- Decision trees: navigation, forms, and debugging flows that minimize image captures.
- Optional Gemini Flash: for true visual analysis,
scripts/process-screenshot.jssends one image to Gemini Flash (~$0.001) instead of round-tripping a large screenshot.
# one-time: configure the Gemini API key (read from
# ~/.config/chrome-devtools-optimizer/config.json — never hardcoded)
node scripts/process-screenshot.js <image-path-or-base64>
See agents/chrome-devtools-optimizer.md for the
full decision logic.