Journey Test Harness Agent Skill

Journey Test Harness

Orchestrate journey-runner across the selected journey directories. The harness must work on Windows, Mac, and Linux and must never delete an Azure environment merely because it appears in azd env list.

Inputs

• Journeys, default all
• Stack, default Node.js/TypeScript for multi-stack journeys
• Deploy, default true
• Location, default westus
• Concurrency, default 1; increase only when quota, local ports, and API limits allow it

Each deployed journey uses cleanup: after-verification unless the user explicitly requests otherwise.

Step 1: Discover Journeys

Use repository file-search APIs or Node.js fs.readdir() to find journeys/*/README.md. Do not use ls | sed, shell globs, or platform-specific path parsing.

For each journey, record:

• Folder name and absolute source path
• Journey type
• PLAN.md presence
• Required and optional tools from the journey prerequisite section
• Estimated cost and time
• Screenshot requirement

Filter only against the discovered folder names. Fail early for an unknown requested journey.

Step 2: Suite Preflight

Load journey-runner and run its cross-platform preflight before creating any workspace or Azure resource.

The union of selected-journey requirements may include:

• Node.js LTS or later
• Azure CLI and valid authentication
• azd 1.28.0 or later with auth.useAzCliAuth=true
• GitHub Copilot CLI
• Node.js LTS or later for hooks, tests, and verifiers
• Azure Functions Core Tools v4, Azurite, and Go-based sqlcmd for SmartTodo
• The pinned Playwright package and bundled Chromium for web screenshots
• Xcode 16+ only when SmartTodo iOS execution is requested on Mac

Use .github/skills/journey-runner/scripts/check-prerequisites.mjs with the union of required tools. Missing required tools stop the whole suite before provider registration. Do not install system tools during the suite.

Step 3: Azure Preparation

Check Azure CLI and azd authentication separately. Confirm the intended subscription and location.

Register only providers required by selected journeys. Run provider commands as individual processes with argument arrays, not shell loops. Wait for required registrations before starting the first deployment.

Record the subscription ID and each provider's state in the suite report.

Step 4: Create Suite and Journey Workspaces

Create directories through Node.js filesystem APIs or the active agent's file tools:

<journey-runs-root>/test-suite-<UTC timestamp>/
├── test-report.md
├── screenshots/
└── runs/
    └── <journey>-<UTC timestamp>/

Use path.join() and absolute paths. Don't embed ~/, date, mkdir, or shell-specific separators in executable instructions.

Copy PLAN.md with fs.copyFile() when needed. Generated application code, infrastructure, logs, and secrets stay inside the isolated run directory, never the source repository.

Step 5: Run Each Journey

Invoke journey-runner with:

Journey: <journey-source-path>
Stack: <stack>
Location: <location>
Working directory: <absolute-run-directory>
Cleanup: after-verification

The runner owns per-journey preflight, prompts, local ports, deployment, verification, screenshots, and scoped cleanup.

Run serially by default. Parallel runs require:

• Distinct local ports
• Distinct azd environment names
• Distinct resource groups
• Sufficient regional quota
• Independent log and screenshot paths

Never share an azd environment across journeys.

Step 6: Preserve Evidence

Before removing a run directory, copy these artifacts with Node.js filesystem APIs:

• run-report.md
• issues.md
• screenshot-*.png
• Sanitized deployment and verification logs

Use the mapping below:

| Journey | Web output | Screenshot |

|---|---|---|

| Grafana | GRAFANA_URL | Login page |

| n8n | N8N_URL | Owner-setup or login page |

| Superset | SUPERSET_URL | Authenticated welcome page when credentials are available |

| AIMarket | WEB_URL | Storefront with all images loaded |

| SmartTodo | None | No screenshot on Windows/Linux; simulator screenshot only on a suitable Mac/Xcode host |

Use journey-runner's pinned Playwright Chromium helper. Never use a branded Chrome channel.

Step 7: Guaranteed Scoped Cleanup

Cleanup runs in a finally path after each deployment attempt, including build, deployment, verification, or screenshot failures.

For each journey, delete only:

• Its explicitly recorded azd environment
• Its explicitly recorded resource groups and managed resource groups
• Soft-deleted resources whose names were recorded as owned by that run

Never run a pipeline over every row from azd env list. Other environments may belong to unrelated work.

After azd down --force --purge --no-prompt, verify:

• Recorded resource groups return not found
• No active resource remains with that run's environment tag
• Any owned soft-deleted resource selected for purge is absent
• Unrelated environments and resource groups remain unchanged

If cleanup is incomplete, preserve the workspace and list exact remaining resource IDs in the consolidated report.

Step 8: Workspace Retention

Delete a per-journey working directory only when all of these are true:

• Evidence was copied successfully.
• Azure cleanup was live-verified.
• The user didn't request generated code retention.
• No unresolved cleanup or debugging artifact depends on the directory.

Use Node.js fs.rm(path, { recursive: true, force: true }) on the exact recorded run directory. Never construct a deletion path from untrusted output.

Step 9: Consolidated Report

Write test-report.md directly with file tools or Node.js. Don't build Markdown through a large shell one-liner.

Include:

• UTC start/end and total duration
• Host OS and architecture
• Tool versions
• Subscription and location
• Selected journeys and stack
• Build, deployment, verification, screenshot, and cleanup status per journey
• Preserved artifact paths
• Exact blockers and remaining resource IDs

Example result table:

| Journey | Build | Deploy | Verify | Screenshot | Cleanup | Result |

|---|---:|---:|---:|---:|---:|---:|

| Grafana | PASS | PASS | PASS | PASS | PASS | PASS |

| n8n | PASS | PASS | PASS | PASS | PASS | PASS |

| Superset | PASS | PASS | PASS | PASS | PASS | PASS |

| AIMarket | PASS | PASS | PASS | PASS | PASS | PASS |

| SmartTodo | PASS | PASS | PASS | N/A on Linux/Windows | PASS | PASS |

A journey can't receive PASS when cleanup was requested but not verified.

Failure Policy

• Missing prerequisite: stop before Azure work.
• Build failure: one evidence-based repair, then mark FAIL.
• Deployment failure: collect diagnostics, then run scoped cleanup.
• Verification failure: retry only documented transient cases, then mark FAIL.
• Screenshot failure: preserve logs and continue to cleanup; mark PARTIAL only if all functional checks passed.
• Cleanup failure: mark FAIL and preserve the run directory.

Safety Rules

• Never install system packages or privileged emulation automatically.
• Never delete all azd environments.
• Never broaden cleanup scope after a failed deletion.
• Never expose credentials, cookies, authorization headers, connection strings, or generated secrets.
• Never claim Mac, Windows, iOS, or browser verification that didn't actually run.