qa-start — скилл для Claude Code: что делает и как установить

---
name: qa-start
description: >-
  Sequenced launcher that bootstraps QA on a project with no QA in place. Chains
  qa-project-context → test-strategy → test-planning in one guided run, then executes
  available checks and records evidence. Use when: "set up QA on a new project," "QA from scratch," "no QA
  exists yet," "/qa-start." Not for: onboarding a QA engineer to an existing team with
  existing tests — use qa-project-bootstrap. Not for: "which skill do I use" with no
  clear match — use qa-do.
  Related: qa-project-context, test-strategy, test-planning, qa-project-bootstrap, qa-do.
license: MIT
compatibility: Cross-tool skill instructions. The optional repository CLI requires Node 24; execution needs the project's test tools and any configured target.
metadata:
  author: kindlmann
  version: "2.1"
  category: foundation
  argument-hint: "optional path to existing repo (e.g. './apps/web') if running from a monorepo root"
---

<objective>
A sequenced launcher for bootstrapping QA where none exists. Inspect the repository,
capture available context, choose the first critical behavior, and execute its checks.
Use `qa-project-context` → `test-strategy` → `test-planning` to keep the decisions explicit,
then deliver test evidence and the next concrete gap alongside the planning artifacts.
</objective>

## When to Use This

Reach for `qa-start` whenever the answer to "where do we start with QA?" is unclear and no QA foundation exists yet:

- A brand-new project with no test infrastructure.
- Joining an existing codebase that has **no QA setup at all** — old code, but quality work was never formalized. Still `qa-start`, not `qa-project-bootstrap` (that one is for joining a team that *already has tests*).
- Rebooting QA after neglect — tests deleted, coverage collapsed, no strategy. You are rebooting QA from the foundation up.

If the codebase already has a real test suite and you are a QA engineer ramping onto the team, use `qa-project-bootstrap` instead.

## Quick Route

Pick your entry point — skip any step whose artifact already exists.

| Situation | Start at |
|-----------|----------|
| No QA at all (no context file, no strategy, no plan) | Step 1 |
| `.agents/qa-project-context.md` already populated | Skip Step 1 — invoke `test-strategy` (Step 2) |
| Context file AND strategy document both exist | Skip Steps 1 and 2 — invoke `test-planning` (Step 3) |
| All three done — what next? | After Step 3 |
| A QA engineer ramping onto an existing team with existing tests | Not this skill — see `qa-project-bootstrap` |

## Step 1: Capture Project Context

**Skill:** `qa-project-context`

Creates `.agents/qa-project-context.md` in your project root. That file records your tech stack, test frameworks, CI/CD pipeline, environments, coverage goals, risk areas, and team structure. Every later skill reads it, so it never asks you the same questions twice.

**What to do:** Invoke `qa-project-context`. Start discovery with the repo and existing
task, then write the file. Record unknown decisions explicitly. Ask only for a missing
answer that changes the next action; continue useful checks when optional context is absent.

**Done when:** `.agents/qa-project-context.md` records observed setup, usable commands,
current scope, and unresolved decisions. Do not invent team structure, targets, or credentials.

## Step 2: Create the Test Strategy

**Skill:** `test-strategy`

Produces a strategy document defining how the project approaches quality: the test pyramid (what proportion of unit, integration, and E2E tests fits your product), entry and exit criteria, tool selection, environment coverage, and quality gates for CI and release.

**What to do:** Invoke `test-strategy` once Step 1 is complete. It reads the context file automatically and will not re-ask anything answered there.

**Done when:** You have a strategy document (e.g. `strategy.md`) covering test-pyramid rationale, tool choices with justification, quality gates for CI and release, and entry/exit criteria per test type. One page of clear decisions beats a sprawling template.

## Step 3: Build the First Test Plan

**Skill:** `test-planning`

Translates the strategy into an actionable plan for the first sprint or release. Maps features to test cases, assigns effort, and decides what gets covered first versus deferred.

**What to do:** Invoke `test-planning` after Step 2. It consumes the strategy from Step 2 and the context from Step 1. Provide the feature list or sprint scope when prompted.

**Done when:** You have a test plan (e.g. `plan.md`) with features mapped to test cases, coverage priorities set, effort estimated, and scope boundaries clear. This is the artifact your team executes against.

## Run the First Checks

Inspect the repository before producing planning documents. Use available context to
make routine choices and keep the initial strategy and plan proportional to the project.
A missing optional context file does not block an authorized local test run.

When the full QA Skills checkout is available, run:

```bash
node <qa-skills>/scripts/qa.mjs init --project <project> --wire-agent
node <qa-skills>/scripts/qa.mjs run --project <project> --goal "Check the critical user flow"
```

`init` creates a project config without overwriting an existing one. Use `--wire-agent`
when the user wants QA integrated into future coding tasks; it adds a bounded workflow
to the project's `AGENTS.md`, preserving existing instructions. Skip init for a configured
project. The host agent must load that project instruction for automatic selection.

Read the selected domain skill and execute its workflow. If there are no tests, author
and run the first test for an observed critical behavior; do not present setup as a pass.
Retain a real `report.html` and `result.json`, including blocked dependencies and unrun
flows. A successful smoke check establishes only the behavior that was exercised.
See the repository's `docs/QA-RUNNER.md` for the runner contract.

## After Step 3

You now hold the foundation: context, strategy, and a first plan. Next actions depend on your stack — this launcher is not the final step.

- **First automated tests:** Use `playwright-automation` or `cypress-automation` to write the first E2E tests against your highest-risk flows.
- **CI integration:** Use `ci-cd-integration` to get tests running on every pull request.
- **Tracking quality:** Use `qa-metrics` once the suite runs to define what health looks like and how to measure it over time.

## qa-start vs qa-project-bootstrap vs qa-do

The single most common routing confusion, answered in one place:

- **`qa-start`** — bootstrap QA where **none exists**. New project, or QA was deleted and you are rebooting it. Sequence: context → strategy → first plan. Output is the QA *foundation*.
- **`qa-project-bootstrap`** — a 30-day ramp for a QA engineer **joining an existing team** with existing tests. Covers team processes, onboarding timeline, and an audit of the test suite already in place. Use it *after* a foundation exists, not to create one.
- **`qa-do`** — last-resort router for "which skill do I use?" when nothing clearly matches. If you already know you have no QA, skip it and run `qa-start`.

A brand-new project with no QA — even on an old codebase — is always `qa-start`, never `qa-project-bootstrap`.

## Pin as a Manual Command

Re-run anytime from the Claude Code `/skills` menu.

To pin this skill as a manual-only command, add `disable-model-invocation: true` to the frontmatter, or set it via `skillOverrides` in `.claude/settings.local.json`.

> Caveat: in some Claude Code builds `disable-model-invocation: true` also suppresses the manual slash command (ref anthropics/claude-code#26251). If you still want to invoke `/qa-start` by hand, set the `skillOverrides` state to `user-invocable-only` instead.

## Related Skills

- **qa-project-context** — Step 1. Captures project setup, tech stack, and quality goals into `.agents/qa-project-context.md`.
- **test-strategy** — Step 2. Defines the testing approach, pyramid, tools, and quality gates.
- **test-planning** — Step 3. Builds the first test plan with features mapped to test cases.
- **qa-project-bootstrap** — go here instead when a QA engineer is ramping onto an existing team with existing tests; this skill creates the foundation, bootstrap onboards onto one.
- **qa-do** — last-resort router when no skill clearly matches; not needed if you already know there is no QA.
- **playwright-automation / cypress-automation** — after the plan, write the first E2E tests against your highest-risk flows.