# Qawolf Onboarding

> Use when the user wants to onboard an application to QA Wolf, get started with QA Wolf, choose their first test, or find the best first flow. Select a useful, achievable onboarding journey, then invoke qawolf-flow-outline for runner exploration, AAA approval, draft publication, an independent run, repair when needed, and activation after success. A specific new-flow request goes directly to qawolf-flow-outline, and a failing existing flow goes to qawolf-flow-maintenance.

- Skill: `qawolf/qawolf-onboarding` (Agent Skill)
- Install (CLI): `npx skillmds@latest add qawolf/qawolf-onboarding`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qawolf/qawolf-onboarding/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: qawolf (https://skillmd.com/u/qawolf)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/qawolf/qawolf-onboarding

---


# Onboarding

Choose the best first flow for this application, then hand it to Flow Outline. Do not maintain a second exploration or implementation workflow here.

This skill is for an application with no coverage yet. A user who named the flow they want goes straight to [Flow Outline](../qawolf-flow-outline/SKILL.md), and one whose existing flow is failing goes to [Flow Maintenance](../qawolf-flow-maintenance/SKILL.md).

## Choose the first flow

Apply the shared [connection checks](../qawolf/SKILL.md#start-here) and [data and write safeguards](../qawolf/SKILL.md#protect-data-and-confirm-writes). Reuse known context and skip sign-in when already connected. Ask for the application URL or goal only if missing; use the client's ask-user tool when available. Never request credentials in chat.

If the user already named the flow, use that choice and invoke Flow Outline immediately. Do not run a separate selection exercise or require the user to repeat the goal.

Otherwise, recommend one candidate using the app's purpose, the user's priorities, available test access, and known constraints. Prefer a core user outcome with a short repeatable path, a visible success state, and little destructive setup or cleanup. Login is a candidate, not the automatic choice. Avoid payment, messaging, or customer-data changes unless explicitly approved.

State why the candidate is useful and achievable. Treat it as a hypothesis until runner exploration verifies it. If essential product context or business priorities are missing, use the ask-user tool for that gap, then select one concrete candidate before invoking Flow Outline. Do not replace browser exploration with source inspection or ask the user to map the application for you.

## Invoke Flow Outline

Activate `qawolf-flow-outline` through the client's skill tool, using the registered plugin-qualified name where required. If the client cannot invoke skills, read and follow [Flow Outline](../qawolf-flow-outline/SKILL.md) directly. Loading a skill does not delegate to the remote QA Wolf agent.

Pass the application target, proposed or user-selected journey, selection rationale, known workspace and environment, approved access references, permission boundaries, prior observations, and unresolved gaps. Do not pass secret values or invent missing context.

Flow Outline owns all remaining work: runner computer-use exploration, independent discovery, ask-user questions for gaps, AAA presentation, approval for the full lifecycle, `agent_send`, draft publication, independent runs, repairs, and readiness. It may refine the candidate based on observations before presenting the outline. Do not call `agent_send` yourself or launch a second creation session after the handoff. Selecting a candidate is not approval to create it. Finish the approved batch, then report its results; another onboarding journey needs its own visible outline and approval.