High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and real-time dashboard.
Tests
This directory contains benchmark tasks and supporting infrastructure for measuring AI agent performance with PinchTab and other browser-control surfaces.
Directory Structure
tests/
├── benchmark/ # Benchmark task definitions and results
│ ├── group-XX.md # Task groups (one file per group)
│ ├── index.md # Task index
│ └── results/ # Generated reports
│
├── optimization/ # Optimization task definitions (future)
│
└── tools/ # Benchmark infrastructure
├── runner/ # Go benchmark runner
├── scripts/ # Shell wrappers (pt, ab, baseline.sh, etc.)
├── fixtures/ # Test HTML pages
├── docker/ # Smoke test Dockerfiles
└── config/ # PinchTab configuration variants
Quick Start
# From repo root:
# Deterministic baseline verification (no API keys)
./dev opt baseline
# Agent-driven benchmarks
ANTHROPIC_API_KEY=... ./dev bench pinchtab
ANTHROPIC_API_KEY=... ./dev bench agent-browser
Lanes
| Lane | Command | Description |
|---|---|---|
| baseline | ./dev opt baseline |
Deterministic shell verification |
| pinchtab | ./dev bench pinchtab |
Agent-driven via PinchTab CLI |
| agent-browser | ./dev bench agent-browser |
Agent-driven via agent-browser CLI |
Task Groups
Tasks are organized into groups in benchmark/group-XX.md:
- Group 0: Setup & diagnosis
- Group 1: Reading & extracting content
- Group 2: Search functionality
- Group 3: Form interactions
- Groups 4+: Advanced scenarios (SPA, auth, modals, etc.)
Reports
Generated reports appear in benchmark/results/:
*_benchmark_YYYYMMDD_HHMMSS.json- Raw JSON data*_summary.md- Human-readable summary*_commands.ndjson- Command trace
See tools/README.md for infrastructure details.