Skip to content

Configure GitHub Actions

Status: ⚠️ It works: it triggers on its own when a PR has the feature label and it comments the result. But it doesn't automatically open the PR with the results: it pushes the branch, and you have to open the PR by hand.

Value in qa-stack.yaml stack.pipeline_provider: github
Entry file .github/workflows/qa.yml (or .github/workflows/neo-ci-jobs.yml if qa.yml already existed)
NEO files .github/workflows/neo-ci-jobs/neo-ci-manager.yml
How it's triggered Pull Request with the feature label
Image registry.gitlab.crowdaronline.com/lippia/products/neo/crowdar-engine-mvp:c4cd2740-playwright (private registry)

1. What lippia-neo-ci-bootstrap generates

File What it contains
.github/workflows/qa.yml The entry point: it triggers on PRs and calls the manager with uses: + secrets: inherit
.github/workflows/neo-ci-jobs/neo-ci-manager.yml A reusable workflow with the NEO jobs

The generated entry point:

name: QA — playwright
on:
  pull_request:
    types: [opened, labeled, reopened, synchronize]
jobs:
  neo-ci:
    uses: ./.github/workflows/neo-ci-jobs/neo-ci-manager.yml
    secrets: inherit

If you already had your own .github/workflows/qa.yml, it isn't touched: the NEO entry point is written as .github/workflows/neo-ci-jobs.yml. GitHub runs each workflow separately, so there's nothing to merge.


2. The jobs

Job What it does
initial-setup Detects the framework from qa-stack.yaml and installs dependencies (and leaves the cache ready)
feature-e2e Clones the skills, writes .env.tests, reads the PR (neo event), runs the feature-new-end-to-end plan, and pushes the results to a branch
publish-report Comments the result on the PR. It always runs, even if feature-e2e fails

All three run only if the PR has the feature label: when it's opened, when the label is added, when it's reopened, and on every push (synchronize).


3. Step by step

Step 1 — Generate the files

With qa-stack.yaml already created (pipeline_provider: github), ask the assistant:

Generate the project's CI with lippia-neo-ci-bootstrap

Review the generated files and commit them.

Step 2 — Add the secrets

In GitHub: Settings → Secrets and variables → Actions → New repository secret.

Secret What it's for
CROWDAR_REGISTRY_USER User to download the crowdar-engine-mvp image from Crowdar's GitLab registry
CROWDAR_REGISTRY_TOKEN That user's token, with read permission on the registry
SKILLS_REPO_TOKEN Token with read access to the neo-releases repository, which is on GitLab
ANTHROPIC_API_KEY Claude models
OPENAI_API_KEY OpenAI models, if the plan uses them
CROWDAR_CI_TOKEN Commenting on the PR and pushing the results branch
APP_USER / APP_PASSWORD The app's test user
APP_BASE_URL URL of the app under test

The skills and the image live on Crowdar's GitLab, not on GitHub. That's why you need GitLab tokens (CROWDAR_REGISTRY_* and SKILLS_REPO_TOKEN) stored as GitHub secrets. Ask the NEO team for them.

PRs from forks: GitHub doesn't pass secrets to workflows triggered from a fork. In that case the jobs fail because of empty credentials: NEO is designed for PRs from the same repository.

Step 3 — Create the label

In GitHub: Issues → Labels → New label → name feature. (The condition looks for exactly feature.)

Step 4 — Test it

  1. Open a test PR and add the feature label.
  2. In the PR's Checks tab you should see Initial Setup, Feature E2E, and Publish Report.
  3. When it finishes, the PR gets a comment with the result. If files were generated, the feature-e2e log shows the branch that was pushed: open the results PR by hand from that branch.

4. If your user stories come from Jira or Azure DevOps

⚠️ Current limitation: the generated feature-e2e job only passes ANTHROPIC_API_KEY, OPENAI_API_KEY, and CROWDAR_CI_TOKEN to the engine. It doesn't pass the Jira or Azure DevOps variables, so with user_stories_storage.type: cloud the MCP server can't authenticate.

To fix it:

  1. Add the Jira or Azure DevOps secrets in Settings → Secrets and variables → Actions.
  2. Add them to the env: block of the feature-e2e job in neo-ci-manager.yml:
    env:
      # ...the ones already there...
      JIRA_URL: ${{ secrets.JIRA_URL }}
      JIRA_USERNAME: ${{ secrets.JIRA_USERNAME }}
      JIRA_API_TOKEN: ${{ secrets.JIRA_API_TOKEN }}
      CONFLUENCE_URL: ${{ secrets.CONFLUENCE_URL }}
      CONFLUENCE_USERNAME: ${{ secrets.CONFLUENCE_USERNAME }}
      CONFLUENCE_API_TOKEN: ${{ secrets.CONFLUENCE_API_TOKEN }}
      # or, for Azure DevOps:
      # AZURE_DEVOPS_ORG_URL: ${{ secrets.AZURE_DEVOPS_ORG_URL }}
      # AZURE_DEVOPS_AUTH_METHOD: pat
      # AZURE_DEVOPS_PAT: ${{ secrets.AZURE_DEVOPS_PAT }}

If you run lippia-neo-ci-bootstrap again, make sure this change isn't lost.


5. Common problems (and what to do)

What you see Why it may happen What to do
The jobs show up as skipped The PR doesn't have the feature label Add the label
Error response from daemon: unauthorized when starting the container CROWDAR_REGISTRY_USER / CROWDAR_REGISTRY_TOKEN are missing or don't have access to the registry Check the secrets
Clone crowdar-qa-skills fails SKILLS_REPO_TOKEN is missing or doesn't have read access to the repository Check the secret
The credentials arrive empty The PR comes from a fork, or secrets: inherit is missing from the entry point Use branches from the same repository; check qa.yml
The results PR isn't opened Known limitation: GitHub has no equivalent to GitLab's merge_request.create Open the PR by hand from the branch the job pushes
The plan can't read the user story from Jira / Azure The job doesn't receive those variables See section 4
It can't comment on the PR CROWDAR_CI_TOKEN doesn't have permissions on the repository Check the token