AgentKit
DocsKitsCLI ReferenceDesktop App

Skills

Verify Marketing interfaces and workflows with ak:test

Run a bounded UI or workflow test, preserve authentication and live-system boundaries, and produce screenshots or step evidence without implementing fixes.

Use ak:test to verify a website interface or a Marketing component workflow. The UI route plans browser checks and produces a report with screenshots. The workflow route scans available components, runs selected steps, and records a human Pass, Fail, or Skip decision at each checkpoint.

Choose ak:test for verification

Use ak:test when

  • You need visual, accessibility, responsive, navigation, form, or usability evidence for a site you are authorized to test.
  • You want step-by-step validation of a Marketing command, Agent, or Skill.
  • You need to scan project components and generate candidate test scenarios.
  • You want a concise report that keeps failures visible and stops before fixes.

Choose another workflow when

  • You need to diagnose and implement a correction. Finish this test report first, then use a debugging or implementation workflow with the evidence.
  • You need code-level unit or integration tests without the Marketing UI or workflow protocol.
  • You do not control the target site or lack permission for automated browsing, protected routes, forms, performance checks, or security checks.
  • You need to claim broad browser compatibility from one available browser or one viewport. Record only the environments actually exercised.

Prepare the target safely

Before you start:

  • Confirm Marketing Kit is installed for the current runtime and scope.
  • Provide ui plus an authorized URL, or workflow plus a supported workflow target.
  • Define pages, viewports, user flows, exclusions, expected behavior, and the stop condition.
  • Prefer staging, fixtures, or non-production data. State whether any form may submit or any account state may change.
  • For protected routes, use an approved session or test account and keep credentials out of prompts, screenshots, reports, logs, and committed files.
RuntimeInvocationAvailability boundary
Claude Code/ak:test ...Can route to packaged browser and component-testing guidance when the required tools are available.
Cursor/ak:test ...Uses slash invocation; browser, Agent, and screenshot capabilities still depend on the active setup.
Codex$ak:test ...Uses native Skill discovery; available browser tools and delegated testing remain runtime- and permission-specific.

Run the Skill

The first positional word selects ui or workflow. This example keeps the UI test read-only and names the intended viewport:

/ak:test ui https://staging.example.com/pricing --mobile — inspect only; do not submit forms, change account state, or implement fixes

Select a route deliberately

RouteInputExpected outcome
ui <URL> [options]Authorized site and optional settings such as --headless, --mobile, or --authA scoped test plan, executed browser checks, screenshots, findings, and a Markdown report
workflow <target>youtube, content, email, brand, or allComponent scan, ordered steps, captured output, manual Pass/Fail/Skip decisions, and summary counts

The component scanner supports --json, --scenarios, and --type <commands|agents|skills|workflows|all>. It scans .claude/commands/, .claude/agents/, .claude/skills/, and .claude/workflows/; a missing directory produces an empty category rather than proving the component exists elsewhere.

Understand the UI stages

  1. Confirm authority and scope. The run records the URL, routes, data, viewport, authentication boundary, and prohibited mutations.
  2. Establish browser state. Use the available browser Skill or approved Chrome integration. Protected-route state is injected only through a secure, project-native mechanism you approve.
  3. Discover within bounds. The run inventories reachable pages, components, forms, navigation, and endpoints without expanding to unrelated domains.
  4. Create the test plan. Checks cover only the requested functions, accessibility, responsive behavior, usability, or other named concerns.
  5. Execute and capture evidence. Record environment, action, expected and observed behavior, and screenshots in one report directory.
  6. Report without fixing. Summarize findings and recommendations, then stop.

Understand the workflow stages

  1. Scan the installed project components and generate scenarios when requested.
  2. Load the selected workflow definition.
  3. Display each action, input, and verification checklist.
  4. Execute the available component and capture its output.
  5. Ask a person to record Pass, Fail, or Skip; do not convert missing evidence into a pass.
  6. Report totals and details for every step.

The bundled workflow definitions include historical command names. Marketing Kit currently ships no command catalog, so verify each referenced action is actually present in the project before running it. An absent action is a limit or skip, not a successful test.

Protect authentication and live systems

Test access is not publication or mutation authority

Do not paste cookies, bearer tokens, session storage, or other credentials into the prompt or report. Authorization to view a protected route does not permit submitting a form, sending a message, changing a live account, creating load, probing security controls, spending provider credits, or publishing content. Approve those effects separately and use least-privileged test data.

If authentication is required, record only the type of approved test state and how to reproduce it safely; redact values. Clear temporary state when requested and avoid screenshots that expose customer or account data.

Verify outputs and evidence

A complete UI report should include scope, environment, pages and viewports tested, result per check, reproducible steps, screenshot references, severity, limitations, and unresolved questions. It should also state what was not tested and confirm that no fix was implemented.

A complete workflow report should include the selected workflow, discovered component paths, output per step, the human decision for each step, and exact Pass, Fail, and Skip totals. Preserve failed and skipped evidence for the next diagnostic workflow.

Troubleshoot and respect limits

SymptomSafe next step
Browser tooling is unavailableStop the UI route or switch to an explicitly approved available browser; do not claim the checks ran.
A protected route redirects to loginRe-establish an approved test session without exposing credentials.
A workflow action is absentRecord Skip or a blocking limit and verify the installed project catalog.
Screenshots contain sensitive dataDo not publish or commit them; redact or recapture with safe fixtures.
A test finds a defectFinish the evidence report and ask before starting any fix.

The Skill cannot prove cross-browser behavior, accessibility conformance, performance, security, or production safety beyond the exact tools and checks run. Continue with Marketing Kit, Marketing Kit Agents, or the appropriate diagnostic workflow after the report is accepted.