Plan Claude Code setup with ak:claude-code

Use ak:claude-code for a focused question or change plan about Claude Code. The Marketing Skill routes the topic to bundled references for installation, custom Skills, slash commands, MCP, Hooks, plugins, settings, IDE integration, CI/CD, enterprise controls, advanced features, troubleshooting, APIs, and operational practices.

Invoking this Skill does not prove that every referenced command or setting is current. Claude Code and its service surfaces evolve independently of the Kit. Record the installed version and reconcile packaged guidance with the current official contract before applying a change.

Choose ak:claude-code for product guidance

Use ak:claude-code when

• You need to understand or plan Claude Code installation and authentication.
• You are designing or reviewing Skills, slash commands, Hooks, plugins, MCP servers, settings, or IDE integration.
• You need a CI/CD or enterprise deployment review with security, access, and data boundaries made explicit.
• You are troubleshooting an observed authentication, connection, permission, configuration, or performance issue.

Choose another workflow when

• The task is about AgentKit installation or Kit lifecycle rather than Claude Code itself.
• You need to execute an existing MCP tool; use ak:use-mcp.
• You need current provider facts that are absent from the installed version and verified documentation.
• You want an installation, account, configuration, CI, or enterprise mutation but have not approved its exact target and effects.

Prepare the product context

Before you start:

• Complete Onboarding and install Marketing Kit for the runtime and scope you use to invoke the Skill.
• Distinguish the invoking runtime from the subject: you can ask the Skill from Claude Code, Cursor, or Codex, but the guidance is about Claude Code.
• Record claude --version, operating system, installation method, delivery scope, relevant project instructions, and the observed symptom.
• Identify whether the answer needs current official verification. Do not rely on bundled model IDs, prices, limits, install methods, settings, or platform features without it.
• Redact API keys, access tokens, cookies, SSO assertions, connection strings, private logs, and customer data.
• For a proposed change, state whether reading config, editing files, installing software, connecting to a service, or changing an account is authorized.

See Runtime adapters for the distinction between Skill projection and runtime parity.

Start with a read-only audit

/ak:claude-code Review the Claude Code Skills, Hooks, and MCP setup in this project against the installed Claude Code version. Report stale or risky configuration and propose a minimal patch, but do not edit files, install software, connect services, or expose secrets

/ak:claude-code Review the Claude Code Skills, Hooks, and MCP setup in this project against the installed Claude Code version. Report stale or risky configuration and propose a minimal patch, but do not edit files, install software, connect services, or expose secrets

$ak:claude-code Review the Claude Code Skills, Hooks, and MCP setup in this project against the installed Claude Code version. Report stale or risky configuration and propose a minimal patch, but do not edit files, install software, connect services, or expose secrets

Follow the guidance stages

1. Classify the topic. Select the smallest relevant area: setup, Skill, command, MCP, Hook, plugin, settings, IDE, CI/CD, enterprise, API, advanced feature, or troubleshooting.
2. Collect versioned evidence. Record the installed version, exact symptom, scoped config, and provider or platform boundary. Read only the relevant bundled reference.
3. Reconcile changing facts. Check current official behavior before claiming a command, setting, model, API, limit, price, or feature is available.
4. Separate advice from action. Produce a diagnosis or minimal change plan first. Name every local, account, network, CI, or enterprise mutation.
5. Apply only approved scope. If authorized, make the narrow change while preserving unrelated configuration and secret handling.
6. Validate and report. Use the installed version's diagnostics or a bounded reproduction, then report evidence, remaining uncertainty, and rollback.

Keep configuration and services under approval

Never commit secrets or place real values in example config. Prefer approved environment or secret-manager references with least privilege. Installing or updating Claude Code, adding a plugin, registering an MCP server, enabling a Hook, modifying CI, or changing SSO and access policy are separate mutations; the Skill does not authorize them by being invoked.

Verify the evidence

A complete response should provide:

• The Claude Code version, environment, and topic assessed.
• The bundled reference used and any fact that required current verification.
• Observed configuration or error evidence with secret values removed.
• A minimal recommendation or patch scoped to the exact target.
• Commands or checks actually run and their results.
• Every local, service, account, CI, data, or spend effect.
• Rollback guidance and unresolved version or platform uncertainty.

Troubleshoot and understand limits

The packaged references are a point-in-time knowledge base, not a guarantee of current Claude Code commands, model names, APIs, pricing, or enterprise features. The Skill does not install, authenticate, configure, or operate Claude Code unless those actions are explicitly approved and verified.