AI Quality Toolkit
Shift-left quality that an agent can actually run.
Test suites rot, flakiness hides real regressions, and AI coding agents ship plausible-but-wrong changes when nothing checks them.
Reusable scaffolding that pushes quality left: stable page-object patterns, governance checks that fail the build on unsafe changes, and AI-assisted triage that clusters failures instead of drowning you in them.
Framework-agnostic TypeScript packages plus Playwright fixtures and page objects. Governance runs as CLI checks in CI; failure analysis pairs Currents run data with a triage pass that groups flakes by root cause.
Quality engineering lead. TypeScript, Playwright, Currents, GitHub Actions, and a set of agent skills for review and triage.
- A shared page-object and fixture library extracted from real suites.
- A weekly shift-left scan that opens draft PRs for recurring bug patterns.
- Selector-stability and silent-failure reviewers as reusable agents.
- One investigator per failing shard, in parallel, beats serial debugging.
- Verify every audit finding against source before proposing a fix.
- Reusable artifacts outlast one-off answers.
The toolkit is the engineering spine behind everything else here: the discipline that lets a small team, or a single person with agents, ship changes without lowering the bar.
It treats flakiness, lint, and coverage gaps as first-class defects, and it assumes an AI agent is one of the contributors - so the checks have to be machine-runnable, not just documented.