AI WebGPU Lab Agent

Browser Agent Local Readiness

`exp-browser-agent-local` records a deterministic browser task-automation baseline before real local planner, tool routing, DOM policy, and browser-control runtime wiring land.

The harness fixes local task deck, tool catalog, step trace, task completion, intervention handling, and schema-aligned agent metrics in one readiness surface.

Run Controls

Probe capability first, then run the deterministic browser task deck to export agent success, step latency, tool success, and intervention metadata.

Workspace Snapshot

Checks

  • Keep workflow id, task deck, tool set, and step plan fixed.
  • Record task success, average step latency, tool-call success, intervention count, and fallback metadata before real browser agent wiring lands.
  • Use this readiness surface as the seed input for later agent latency benchmark and voice-agent app experiments.

Task Deck

Draft Output

No agent run yet.

Metrics

Environment

Activity Log

    Schema-Aligned Result Draft

    {
      "status": "pending"
    }

    What This Unlocks

    • First browser agent raw JSON and screenshot capture path
    • Stable task, tool, and intervention metadata for later planner/runtime integration
    • Reusable readiness fixture for `bench-agent-step-latency` and `app-voice-agent-lab`