Agent · QA (retired)
The QA agent is retired. Test generation is a direct skill invocation, and the reviewing half of QA's old job is covered — much more thoroughly — by the layered code review.
Where each duty went
| Old QA duty | Replacement | Notes |
|---|---|---|
[QA] Generate API + E2E tests | skill:e11-qa-generate-e2e-tests | Same skill, invoked directly; detects your test framework from package.json |
| Post-implementation quality pass | skill:e11-code-review | Six default review layers: adversarial-general, edge-case hunter, verification-gap, reachability, acceptance auditor, alloy citation auditor |
| "Did we verify this?" instinct | e11-review-verification-gap layer | Finds changed behavior that could regress without verification catching it |
| "Would this actually work in prod?" instinct | e11-review-reachability layer | Finds activation steps nobody performs, proof aimed at an artifact the runtime never reads, and capabilities nothing registers |
Why
A persona that owned a single command was ceremony. The layered review architecture replaced "ask QA to look at it" with data-driven review layers that run as parallel finder subagents inside e11-code-review and e11-dev-auto — including unattended runs where no one is around to address an agent.
If an older install had a qa.customize.yaml, the upgrade reports it as unmappable — see Upgrading Foundry.
See also
Phase 4 — Implementation · Run an adversarial review · Agent quick-reference.