Exploratory QA with AI Agents: Building a Site-Agnostic Harness

explore-qa is a small Bun runner I built to drive AI agents through exploratory QA sessions. You give it a charter (“explore the basket flow with a mobile viewport”), pick an agent (Claude, Codex, or Copilot), pick a browser CLI (agent-browser or playwright-cli), and it runs the session. You get back a Markdown report shaped like a real tester’s notes: SBTM sections, PROOF debrief, screenshots, full JSONL session log. About 500 lines of TypeScript plus a folder of Markdown. Site-agnostic by construction. Repo: [github.com/alexanderop/explore-qa](https://github.com/alexanderop/explore-qa). Use it as a template, clone it, and run the `/onboard-site` skill to wire it up to your own site. You'll need one of [Claude Code](https://docs.claude.com/en/docs/claude-code/overview), [Codex CLI](https://developers.openai.com/codex/cli), or [GitHub Copilot CLI](https://docs.github.com/en/copilot/concepts/agents/about-copilot-cli) installed to drive the runs. git clone the repo, bun install, bun…