| name | qa |
| description | Use when running an agentic QA pass on the AIdome Endpoint Switchboard extension — test planning, bug hunting, edge-case analysis, or pre-release verification. Wraps the gem-team-inspired QA agent pipeline (research → plan → generate → run → fix) plus adversarial review. |
Agentic QA Skill
A unified entry point for the QA pipeline. Use this skill when you need a
systematic quality pass rather than a single test tweak.
When to invoke
- Adding or substantially changing an adapter, profile, or orchestration flow
- Before cutting a release (Phase 4 / 5 of the extension lifecycle)
- Investigating a flaky or regression-prone test
- Auditing agent files themselves for governance gaps
- Any time a reviewer asks "did we cover the edge cases?"
Agent roster (see .github/references/qa-workflow.md for attribution)
| Agent | Role |
|---|
qa | Adversarial QA specialist — test plan + bug hunt |
critic | Challenges assumptions, finds logic gaps |
critical-thinking | Pre-implementation questioning (no edits) |
test-researcher | Analyze codebase for testability |
test-planner | Phased test implementation plan |
test-generator | RPI pipeline driver for generating tests |
test-fixer | Fix compile / test failures |
test-runner | Run tests, report results |
agent-governance-reviewer | Reviews agent configs for governance gaps |
test-engineer | Repo-native Vitest specialist (integration point) |
Recommended pipeline
1. test-researcher → inventory source files, existing tests, framework (Vitest)
2. test-planner → phased plan covering adapters, profiles, orchestrators
3. critic → critique plan: assumptions, edge cases, over-engineering
4. test-generator → generate tests phase by phase (delegates to test-engineer
for repo-specific Vitest + vscode mock patterns)
5. test-runner → npm test; report pass/fail
6. test-fixer → fix compile / assertion errors; loop ≤ 3 times
7. qa → exploratory / adversarial pass on merged result
8. agent-governance-reviewer → if any .agent.md file was touched
Local invariants the QA pipeline must enforce
npm run lint — zero errors; no console.log in src/npm run compile — zero TypeScript errors (strict mode)npm test — all unit tests + pre-release validation tests pass- All credentials only in
vscode.SecretStorage
- Every adapter
configure() creates a timestamped backup before writing
- URL scheme allowlist enforced; profile names sanitized
- Redaction utility applied before any log of sensitive values
If the QA pipeline cannot satisfy any of the above, escalate to the
orchestrator; do not weaken the tests or skip them.
Outputs
- Test plan in the PR description or a scratch file
- New/updated
test/** files following testing.instructions.md
- Findings report with Severity (Critical / High / Medium / Low)
- Go/no-go recommendation for release (Phase 5 gate)
See also
.github/references/qa-workflow.md — gem-team alignment + attribution.github/instructions/testing.instructions.md — Vitest conventions.github/agents/orchestrator.agent.md — orchestration contract
