Meet Agatha — our QA Manager

Agatha is an agentic employee: a named role with a defined job, a bounded authority ceiling, and an audit trail. She works alongside Alfred on the dev side, and she owns the Aegis quality, safety and compliance layer for landing pages.

She speaks to you inside the tools you already use — Claude and Slack — through the Aegis MCP. There's no new dashboard to learn.

Looking for the mechanics — how runs are triggered, what the guardrails are, how to triage a failure? See the QA Runner Operations Guide.

What she owns

Test suites are Markdown specifications, held in the Aegis database as a single source of truth. Because a suite is prose rather than code, the same test can run from three places without being rewritten:

  • Aegis Monitor — every hour
  • CI — every time a developer changes the code
  • The build itself — tests written as the page is being built

The authority line

The rule that makes an autonomous QA agent safe to point at production:

Agatha may propose. Only a human may activate.

She can draft tests, snapshot versions, dry-run in a real browser, classify failures and quarantine routes. She may never make a baseline live on her own — not even for a change she's confident is benign. New tests land as preview until a person promotes them.

Why the failure triage matters

A failing test has two opposite causes, and confusing them is the one genuinely dangerous mistake:

  • Drift — the page changed but still works. The test is stale; refresh the baseline (with human approval).
  • Breakage — the flow itself is broken. The site is wrong. Refreshing here would rewrite the baseline to match a broken page, turn the test green, and hide a real bug.

So "test failed → refresh the test" is never wired up directly. Every failure is classified first, and anything ambiguous escalates rather than auto-resolving.

Where this is heading

  1. A developer or an AI builds a page
  2. It's added to Aegis
  3. Tests are created as it's built — in Admin (route Options → QA Test), in Claude, or wherever suits
  4. Anyone in the company can run a test on any page, any time, via Claude
  5. Pages are living things, so the whole team can safely update tests via Claude
  6. Tests run on every code change
  7. Tests run every hour
  8. Failures are reported instantly in Slack

What this changes for people

The manual version of this job is someone running a check on every page, every day. With Agatha, the running is automated and the human role moves up: watching what failed and fixing it. Less manual checking, more time on what actually broke.

That shift is the point of the agentic-employee model — not replacing the person, but moving them from executing the checks to managing the system that executes them.