Verification Checklist

Work through this checklist after running marko devai:install to confirm the full integration is working. Each section is independent — if one step fails, consult Troubleshooting before continuing.

1. Package installation

• composer show marko/devai returns a version without error
• marko list includes devai:install, mcp:serve, lsp:serve, and indexer:rebuild in the output

2. devai:install output

Run marko devai:install and verify:

• The installer reports at least one agent detected
• No error messages are printed (warnings are acceptable)
• The installer reports writing at least one agent guidelines file
• Re-running marko devai:install a second time completes without error (idempotent)

3. Agent guidelines file

Choose one agent and verify its guidelines file:

Claude Code:

• CLAUDE.md exists in the project root and references @AGENTS.md
• AGENTS.md exists in the project root and includes a ## Package Guidelines section
• .claude/settings.json registers extraKnownMarketplaces.marko and enables the marko-mcp@marko plugin
• In Claude Code, run /mcp — the marko-mcp server appears (it is provided by the marko-mcp@marko plugin, not by claude mcp add)

Codex:

• AGENTS.md exists in the project root
• .agents/skills/ contains skill files

Cursor:

• .cursor/rules/marko.mdc exists and contains alwaysApply: true
• .cursor/mcp.json exists and uses the mcpServers key

Copilot:

• .github/copilot-instructions.md exists
• .vscode/mcp.json exists and uses the servers key

Gemini CLI:

• GEMINI.md exists in the project root
• .gemini/skills/ contains skill files

Junie:

• junie/guidelines.md exists (under junie/, not .junie/)
• junie/skills/ contains skill files

4. MCP server

• marko mcp:serve starts without error (blocks waiting for stdin — Ctrl+C to exit)
• The agent’s MCP configuration file exists and references marko mcp:serve

Invoke the MCP server from your chosen agent:

• Ask the agent: “What MCP tools are available?” — it should list at least find_event_observers, validate_module, list_modules, list_routes, app_info, read_log_entries, and run_console_command
• search_docs appears in the list only if a docs driver (marko/docs-fts) is installed
• query_database appears in the list only if marko/database is installed
• Ask the agent: “Search Marko docs for routing” — search_docs should return at least one result (requires a docs driver)

5. Code index

The MCP/LSP servers lazy-load and auto-rebuild .marko/index.cache on first read, so explicit indexing is optional. To pre-warm or to force a rebuild after large codebase changes:

• marko indexer:rebuild completes without error and prints module/route/config-key counts
• After indexing, asking the agent “List all Marko modules” (which calls the list_modules MCP tool) returns the expected modules
• find_event_observers returns results when given a known event class FQN that has at least one #[Observer] in your codebase

6. LSP server

• marko lsp:serve starts without error (blocks waiting for stdin — Ctrl+C to exit)
• The editor’s LSP configuration references marko lsp:serve
• In a PHP file, type a config getter and place the cursor inside the quotes (e.g., $config->get("|")) — Marko config key completions appear (textDocument/completion)
• Hover over a quoted config key string — markdown hover with type, default, and source location appears (textDocument/hover)
• Go-to-definition on a quoted config key navigates to its definition (textDocument/definition)

7. End-to-end agent task

Perform one real agentic task to confirm the full integration:

1. Open your chosen agent in the project root
2. Ask: “Create a new Marko module called Greet with a single GET route at /greet that returns ‘Hello, world!’”
3. Verify the agent:
   • Creates app/Greet/ with the correct module structure
   • Uses PHP attribute routing (#[Get('/greet')])
   • Does not use any patterns that violate Marko conventions
4. Ask the agent to call the validate_module MCP tool against app/Greet — it should pass with no errors

8. Skills (if applicable)

The skill files devai distributes live under each agent’s skills directory (e.g., .claude/skills/, .agents/skills/, .gemini/skills/, junie/skills/):

• At least one SKILL.md file exists in the chosen agent’s skills directory
• Asking the agent to perform a skill-specific task triggers the correct skill workflow

All checks pass?

Your AI-assisted development setup is complete. Return to AI-assisted Development overview to explore what else you can do.

If any check failed, see Troubleshooting for targeted fixes.