Evidence
System surface
No public screenshot on file—generative preview stands in for the visual layer.
Story arc
How this shipped
Building the tools that build the toolsThree beats: what pressed against the work, how the stack answered, and what changed once it was live.
I · Constraint
The brief
Reproducing bugs reliably is the hardest part of QA. Evidence gets lost, steps get forgotten, and AI assistants have no direct access to QA findings.
II · Build
The craft
Autonomous Playwright-based crawler with 7 parallel detectors. HAR capture freezes the exact network state of a bug. Reproducibility validation confirms the bug before filing. MCP server exposes findings directly to Claude Desktop for AI-powered triage.
III · Proof
What moved after launch
170 tests across 11 files at 85% coverage. MCP integration means Claude can query and reason about bugs directly. HAR replay via Playwright makes every bug reproducible on demand.
Signals
Signals & scale
Engine room
At a glance
Inventory