---
title: GitHub audit runner
published: true
index: 6
tier: free
---

# GitHub audit runner

Run the Comprehensive Shopify Audit on a GitHub-hosted Linux runner. The action ships inside the skill and uses the same immutable Store Skills release. It reads the complete active Check catalog through the project MCP, runs Codex with an anonymous browser, validates report coverage, and uploads `report.json` and `provenance.json` as the **comprehensive-shopify-audit** artifact. Incomplete or invalid reports fail the job; observed Issues are report results and do not fail it.

## Setup

Publishing the bundled runner requires a Glue deployment and installer that support actions shipped inside skills. Deploy Glue and release a compatible installer before publishing a Store Skills release with this task. Existing releases and installed workflows keep their old action until explicitly replaced.

1. Sign in with your Store Skills account on the Glue host. In a private consumer repository, run `glue init-ci brickspacelab/storeskills --site https://glue.fast` using the updated installer. Select a published Store Skills release that bundles this action. For unpublished installer verification, use `node /path/to/glue.fast/scripts/glue-skills.mjs init-ci` with the same arguments.
2. Follow the generated instructions to set `GLUE_INSTALL_TOKEN` and `GLUE_CUSTOMER_TOKEN` as Actions repository secrets. Add `OPENAI_API_KEY` for Codex model usage. GitHub bills runner time to the consumer repository; the OpenAI project pays for model usage.
3. Set the repository variable `STORE_SKILLS_STOREFRONT_URL` to a public HTTPS storefront or deployed preview. Add this input to the generated action's `with` block alongside `openai-api-key`:

   ```yaml
   storefront-url: ${{ vars.STORE_SKILLS_STOREFRONT_URL }}
   ```

4. Review and commit both the generated workflow and CLI helper. The workflow installs the pinned skill before invoking `./.agents/skills/comprehensive-shopify-audit/ci`. It runs for same-repository pull requests and supports manual dispatch after reaching the default branch. Use a deployed preview URL when auditing a PR's storefront changes; the checkout alone does not prove which revision the storefront serves.

To migrate a workflow using the old separate action, move its workflow/helper aside deliberately, generate replacements, compare them, and retain consumer-specific inputs. The CLI preserves existing files. Keep the old workflow available until the new one passes an authorized private-consumer canary. No repository deletion, permission changes or secret rotation is required.

Each workflow run uses one audit from the signed-in customer’s allowance. Re-running the same job for the same storefront reuses that audit.

The action needs read-only repository permissions. Browser evidence uses anonymous navigation, viewport changes, accessibility snapshots, and console/request observations. Authenticated access, analytics, screenshots, cart writes and checkout are unavailable; affected Checks record specific limitations where safe alternate evidence is insufficient. The catalog step receives the customer credential, while Codex receives model access through the official action's API proxy.

Optional inputs `responses-api-endpoint` and `model` select a trusted OpenAI-compatible Responses API provider; supply its credential as `openai-api-key`. A protected Glue deployment can use `vercel-protection-bypass` for catalog access; the earlier skill installation also needs the deployment's protection header.

## Maintain and validate

The runtime and locked dependencies were moved from `BrickspaceLab/storeskills-action` at commit `8fd97b968d85e23dc798996cbf89b9433cd335b6`. Keep the anonymous browser policy, private/same-repository restrictions, secret isolation, report validation and third-party action pins intact. The runner is opt-in consumer CI code; normal skill installation and local audits do not execute it or install its dependencies.

From this directory on Node 24:

```sh
npm ci --ignore-scripts
npm test
```

Use a private consumer repository and a Pro test subscriber for hosted acceptance before listing a release. Local tests and source exports do not establish paid access or live model/browser acceptance.
