QA configuration and infrastructure
These two skills prepare a project to work with NEO. They're used only once at the start
(or when part of that configuration is missing), and they're steps 1 and 2 of the
New Project Bootstrap playbook.
lippia-neo-bootstrap-stack
When to use it
When setting up a project for the QA flow for the first time: it auto-detects the testing
framework, the language, the app type, and the CI provider from the repo, and writes
qa-stack.yaml. It also asks which test management tool the test
cases are exported to (Xray, QMetry, or none) and where bugs are reported. It's also triggered
automatically as a pre-check when lippia-neo-meta-run-user-story-e2e can't find that file, or
when any tool-agnostic skill fails because it's missing.
{
"output": {
"qa_stack_path": "qa-stack.yaml"
}
}
| Input | Type | Description |
|---|---|---|
output.qa_stack_path |
Required | path where the generated qa-stack.yaml will be written (it's the only formally required field; everything else is resolved by automatic detection or by asking). |
It doesn't have a formal list of optional inputs. Instead, it always requires resolving the following by interacting with the person (they're never inferred or left empty):
- the
environmentfields (base_url,app_version,os) andauth; - where the stories come from (
paths.user_stories_storage):local(a directory of.mdfiles) orcloudvia MCP (jiraorazure-devops); - where bugs are reported (
paths.bug_reports_storage):localorcloudvia MCP (jira,azure-devops,github, orgitlab). Withcloudit also asks for the destination depending on the tool:jira_project_key/jira_issue_type,azure_devops_project/azure_devops_work_item_type,github_repo/github_issue_label, orgitlab_project/gitlab_issue_type. TheBUG-*.jsonfiles are always written first tolocal_path(defaultbug-reports/); - the artifact directories (
test_plans_dir,tests_dir,reports_dir,scripted_dir,exploratory_dir), one per question, even if the default seems obvious; - the path of the
env_file(default.env.tests, gitignored); - the test management tool (
test_management.tool:xray,qmetry, or none), itsproject_key, and its credentials (XRAY_CLIENT_ID/XRAY_CLIENT_SECRETorQMETRY_API_KEY), which go only in theenv_file, never inqa-stack.yaml. You can answer "later" and they're left pending; - if applicable,
api_contract(api/hybridapp) andlippia_config_path(lippiaframework).
In addition, on every run it makes sure .mcp.json has the playwright server (exploration
needs it, whatever the framework), offers to load the credentials of the chosen cloud
providers there (gitignored, never shown), and adds .claude, .playwright-mcp, and
execution/evidence.txt to .gitignore.
Trimmed profile with k6 or jmeter
If the framework is k6 or jmeter, the assistant runs a trimmed profile: it only asks for the
framework and the CI provider (if not detected), environment.base_url, and the
performance block (API collection, data_dir, scripts_dir, profiles, time_scale, and the
methods that modify data, mutations, with the probe repetitions). It doesn't ask for
authentication, OS, app version, stories, bugs, artifact directories, or the test management
tool, and it doesn't create .mcp.json. paths.env_file stays at .env.tests without
asking or creating the file. With jmeter it detects the binary (PATH, JMETER_HOME,
standard locations) without asking and stores performance.tool_path; if JMeter or Java is
missing, it warns without installing anything.
Output
| Field | Description |
|---|---|
qa_stack_path |
path of the written qa-stack.yaml. |
detection_mode |
auto / interactive / mixed, depending on how much was inferred vs. asked. |
stack |
test_framework (playwright/playwright-bdd/playwright-cucumberjs/cypress/lippia/pytest/k6/jmeter), language (javascript/typescript/java/python), app_type (web/api/hybrid), auth (google_sso/basic_auth/api_token/none), pipeline_provider (github/gitlab/bitbucket). |
environment |
base_url, app_version, os. |
api_contract |
only if app_type is api/hybrid: spec_path, spec_url, and snapshot_path (initial snapshot of the contract when spec_url was resolved). |
mcp_json |
path, framework_server (always playwright), and cloud_provider_server (cloud server(s) added for stories and/or bugs, or null if everything is local). |
performance |
only with k6/jmeter: performance block (collection, data_dir, scripts_dir, profiles, time_scale, mutations; with jmeter, also tool_path). |
test_management |
tool (xray/qmetry/null), project_key, credentials_loaded, and credentials_pending (key names only, never values). |
Chains with
Step 1 of the bootstrap playbook; step 2 is lippia-neo-ci-bootstrap, which reads pipeline_provider
from this qa-stack.yaml. It's also the prerequisite for lippia-neo-meta-run-user-story-e2e and for the
rest of the tool-agnostic skills (for example, lippia-neo-report-bug reads
bug_reports_storage to know where to publish).
lippia-neo-ci-bootstrap
When to use it
When you need a CI/CD workflow that runs the QA skills on every commit, PR, or
schedule. It reads pipeline_provider from qa-stack.yaml and outputs the matching workflow
file (GitHub, GitLab, or Bitbucket). It always runs after lippia-neo-bootstrap-stack —
it aborts if qa-stack.yaml doesn't exist yet.
{
"config": {
"pipeline_provider": "github",
"test_framework": "playwright"
}
}
| Input | Type | Description |
|---|---|---|
config.pipeline_provider |
Required | CI provider already resolved in qa-stack.yaml; determines which adapter to load. |
config.test_framework |
Required | testing framework already resolved in qa-stack.yaml; it's substituted as a placeholder in the workflow template. |
Performance frameworks
With test_framework: k6 or jmeter this skill doesn't apply: it aborts without generating or modifying any files, with the reason framework_not_supported_by_this_skill. Those frameworks are handled with the performance skills.
Its own jobs are always written encapsulated in a neo-ci-jobs/ subfolder inside the
provider's jobs folder (.github/workflows/, .gitlab/, or .bitbucket/), with a
neo-ci-manager.yml as the entry point. If the repo's CI entry file already exists, it isn't
overwritten: in GitLab only an include: line is added, in GitHub a separate workflow is written
(.github/workflows/neo-ci-jobs.yml), and in Bitbucket —which doesn't allow splitting the pipeline across
several files— a merge is proposed and confirmation is requested before writing. If the framework is
playwright, playwright-bdd, or playwright-cucumberjs, the repo must have @playwright/test declared in its package.json (all three share the Playwright installation: npm ci + browsers).
The active jobs (M4 model) are initial-setup → feature-e2e (feature-new-end-to-end plan)
→ publish-report, and they run only on PRs/MRs with the feature label. In Bitbucket the label filter
can't be expressed in the YAML: the plan is set up as a custom pipeline qa-feature-e2e, which
has to be triggered from outside (the webhook handler for Bitbucket isn't implemented in the engine
yet). In GitLab the template
also includes ui-selector-repair (ui-selector-repair plan) and qa-read-stack, which are triggered by a
push to develop that touches src/components/ or src/pages/.
Output
| Field | Description |
|---|---|
provider |
github/gitlab/bitbucket. |
generated_file |
path of the generated or updated workflow file. |
phase |
always scaffolding: the YAML is generated once from the template and stays committed. The M4 jobs it includes already invoke the real engine (crowdar-engine-mvp); they're not placeholders. |
warnings |
warnings. |
Chains with
Always runs after lippia-neo-bootstrap-stack, as step 2 of the bootstrap playbook.
After that, you need to configure the secrets/variables requested by the adapter's README and commit
the generated files.