Skip to content
ADVISOR

Sheet 0.3.1

The Wayne State app stack, surveyed

ADVISOR

ADVISOR stands for Answers, Directions, Versions, Integrations, Standards, Onboarding and Resources.

What applies, and what's next.

An MCP server for the stack, a bit like Laravel Boost but for Wayne State. Your coding agent connects to it and gets nine read-only tools: CREED's rules and support dates, each project's docs, PROCTOR and Booster's upgrade plan. Ask it “which PHP can I build on?” and it answers from CREED's policy file, with the date and the source, instead of from whatever the model remembers.

composer require --dev waynestate/advisor

PHP 8.3 or newer. ADVISOR is on Packagist: composer require --dev waynestate/advisor:^0.3.

Works alongside WELD. You can add ADVISOR to a WELD app or to any other Wayne State PHP app.

Ask the map Get directions

The ADVISOR mark: a speech bubble holding a gold compass needle.

The magic trick, minus the magic

Ask the map

Type a question you'd ask your coding agent. The map draws the route ADVISOR takes to answer it: which tools it calls, and which project the answer comes from.

Or try:
Field map of the stack Every stack project is a place on the map, with ADVISOR's advising office in the middle and your app at the bottom left. The plotted route is written out in the control card beside the map. Office Hours Pond Prerequisite Ridge Elective Hill N 012 tool calls
An illustration of how your agent uses ADVISOR's tools. This page calls nothing; the real server runs on your machine, beside your app.

Read this before you set off

Legend

ADVISOR has nine tools. These six are the ones you'll use most; the others are owner_of, playbook_lookup and upgrade_path. Every tool is read-only, and every answer names its source.

Tools

list_projects
The stack's projects, WayneUI first, each marked built or proposed.
get_project_status
One project: its state, package, PHP constraint, its floor under today's policy, and its next step.
read_project_docs
A project's README or best practices, paged. For WayneUI, its markup contract too.
policy_lookup
Whether a PHP, Laravel, Node, MySQL, Vue or Inertia version is supported today, from CREED. For Composer, the minimum version.
list_stack_rules
CREED handbook rules with their file and line, filtered by your app's shape, by chapter, or by a few words such as session cookie.
run_proctor_check
Runs your app's own vendor/bin/proctor check against the bundled policy and returns the findings.

Prompts

  • orient_me a first walk round the stack, ending in three next steps
  • plan_upgrade plans the next upgrade hop from the policy, PROCTOR and Booster's plan

Resources

  • advisor://policy CREED's policy file
  • advisor://creed/<chapter> one per handbook chapter
  • advisor://guide/<slug> hand-written guides; first: Connect to Oracle on Citadel

Map symbols

  • your app
  • a tool call
  • where the answer lives

Run it next to Laravel Boost. Boost knows Laravel and ADVISOR knows Wayne State, so they don't overlap. In a Laravel app the same server also runs as php artisan mcp:start advisor.

Add a project's own tools. Squire ships an ADVISOR module. List Waynestate\Squire\Advisor\SquireModule in a modules.json manifest, point ADVISOR_MODULES_FILE at it, and your agent gains squire_doctor and squire_drift, both read-only. The loader takes only what a manifest lists, so nothing sneaks onto the map.

Turn-by-turn

Directions

Three steps and a check-up, the same on Windows, macOS and Linux, because ADVISOR is one PHP command rather than a shell script.

  1. Add it to your app

    Add it as a dev dependency. It won't start in production anyway.

    composer require --dev waynestate/advisor
  2. Print the client config, and read it

    This writes nothing yet. Use your client's name: claude, warriorgpt, cursor or vscode. For codex it prints TOML to paste into ~/.codex/config.toml.

    php vendor/bin/advisor install --client=claude
    {"mcpServers":{"advisor":{"command":"php","args":["vendor/bin/advisor","mcp"],"env":{"ADVISOR_ENABLED":"1"}}}}
  3. Write it, then approve it by name

    Then open the project in your client and approve the advisor server when it asks. Don't commit the file it writes. It starts a program for anyone who opens the project, so install warns you if it isn't in .gitignore.

    php vendor/bin/advisor install --client=claude --write
  4. Get a check-up

    Doctor starts the server the way your client will, as a child process over stdio, and says in plain words what passed. A warn you may not care about; a fail is something to fix. On Windows, its stdio line is the proof that the one command works there too.

    php vendor/bin/advisor doctor
    pass php: PHP 8.4.14 satisfies ^8.3
    pass php-path: Clients will start /usr/bin/php, the PHP running this check.
    pass content: data/ matches its manifest.
    pass client:claude: claude is configured to start ADVISOR.
    warn client:cursor: .cursor/mcp.json is not present, so cursor will not start ADVISOR.
    pass mcp: The server starts and offers 9 tools.
    pass stdio: `php bin/advisor mcp` runs as a child process on Linux (PHP 8.4.14): clean stdout, 9 tools, policy_lookup answered, exit 0 (412 ms).
Where install --write puts it
  • claude.mcp.json
  • cursor.cursor/mcp.json
  • vscode.vscode/mcp.json
  • codexprinted only (TOML)
  • warriorgpt.mcp.json

Every one starts the same fixed command, php vendor/bin/advisor mcp. Run your client from the folder that holds vendor/, or set ADVISOR_PROJECT_ROOT.

Other commands

# serve over stdio (your client starts this)
php vendor/bin/advisor mcp

# check the bundled content against its manifest
php vendor/bin/advisor content:verify

# when the server doesn't show up in your client
php vendor/bin/advisor doctor

# every check, in the advisor repo itself
composer ci

Laravel apps

The Laravel adapter runs on laravel/mcp and is auto-discovered. If you want the settings in your app, publish the config:

php artisan vendor:publish --tag=advisor-config

Read out at the start line

Briefings for your agent

Your coding agent can set ADVISOR up for you. Copy a briefing, or open it directly in OpenCode, Claude Code, Codex or ChatGPT. Each one lists the steps and the limits, and tells the agent to ask you before it guesses.

Briefing 1 · Any PHP app, raw or Laravel

Get started with ADVISOR

The agent installs ADVISOR, shows you the client config before writing it, and asks its first question once it's connected.

You are a coding agent helping a Wayne State University developer connect ADVISOR to the app in this folder. ADVISOR is the MCP server of WELD, the Wayne State application stack: read-only tools that answer from CREED's policy and the stack's own docs, with the date and the source, instead of from memory.

Read first:
- https://advisor.apps.wayne.edu/llms.txt, the index of ADVISOR's pages, each one as Markdown. Work from them, not from memory.
- This app's composer.json and its .gitignore.
- Which coding agent you are. ADVISOR writes the config for claude, warriorgpt, cursor and vscode, and prints TOML for codex. For any other client (OpenCode, for one), print the claude config, translate it by hand, and show me before you write it.

Task:
1. Add ADVISOR as a dev dependency with `composer require --dev waynestate/advisor` (PHP 8.3 or newer).
2. Print the client config and show it to me: `php vendor/bin/advisor install --client=claude`, with your client's name in place of claude. Write nothing yet.
3. When I say yes, run it again with --write, and check that the file it wrote is in .gitignore.
4. If the server does not show up, run `php vendor/bin/advisor doctor` and tell me what it found.
5. Once it is connected, call list_projects, then policy_lookup for PHP, and tell me which PHP this app can build on today, with the date and the source.

Constraints:
- ADVISOR is a dev dependency and refuses to run in production. Never move it into require, and never set ADVISOR_ENV to production.
- Never commit a generated client config: it starts a program on whoever opens the project.
- The command it registers is always php vendor/bin/advisor mcp. Do not change it.
- Give separate commands for Windows (PowerShell), macOS and Linux wherever they differ.
- Ask me before you guess, and before you pass --force.

Output: the config you propose, the commands you ran, the doctor result if you ran it, then your questions.

Or take the Laravel route

Briefing 2 · Laravel, Blade or Inertia

ADVISOR in a Laravel app, beside Boost

Uses the laravel/mcp adapter and adds ADVISOR to your client next to Boost, without touching Boost's entry.

You are a coding agent adding ADVISOR to a Laravel app at Wayne State University, beside Laravel Boost. ADVISOR is the MCP server of WELD, the Wayne State application stack. Boost knows Laravel; ADVISOR knows WSU. Run both.

Read first:
- https://advisor.apps.wayne.edu/llms.txt, then the README it links: Install, Quick start, Laravel apps and Configuration.
- composer.json, composer.lock, config/, .gitignore, .env.example, and any MCP client config already in the project (Boost writes one).

Task:
1. Add ADVISOR with `composer require --dev waynestate/advisor`. Its service provider is auto-discovered, and its Laravel adapter runs on laravel/mcp.
2. If I want the settings in the app, publish them with `php artisan vendor:publish --tag=advisor-config`. They land in config/advisor.php and read the same ADVISOR_* variables.
3. Choose with me how the client starts it: `php vendor/bin/advisor install` for my client, or the Laravel way, `php artisan mcp:start advisor`. Either way it needs ADVISOR_ENABLED=1 in its environment.
4. Add it beside Boost's entry without touching Boost's. Then ask ADVISOR policy_lookup for PHP and Laravel, and list_stack_rules with the permutation that fits this app (laravel-blade, laravel-blade-vue or laravel-inertia-vue).

Constraints:
- Keep Boost. ADVISOR does not replace it.
- Off unless ADVISOR_ENABLED is set, and never in production: keep it in require-dev.
- Never commit a generated client config, and never pass --force over an entry you did not write.
- Give separate commands for Windows (PowerShell), macOS and Linux wherever they differ.
- Ask me before you guess, and before you change anything Boost owns.

Output: the client config with both servers in it, the commands you ran, the rules list_stack_rules returned, then your questions.

Your agent can read this page too. llms.txt lists the pages, llms-full.txt has all of them in one file, and this page has a Markdown twin. The buttons only copy text or open the tool you picked. This page sends nothing back to us.

Hatched on the map, for a reason

Out of bounds

An advisor can tell you what's required. It cannot sit the exam for you, and this one cannot touch your files.

  • Off unless you turn it on. No ADVISOR_ENABLED, no server. production always refuses.
  • Read-only. No tool takes a command, a URL or a free path, and none of them writes a file.
  • No network, no model calls. It serves a bundled copy of the stack's docs, so it works offline.
  • Verified content. At start-up, every bundled file is checked against a SHA-256 manifest. If one doesn't match, the server won't start.
  • Answers are data. Every result is labeled as reference text and cleaned of invisible characters. Pass or fail is PROCTOR's call in CI.
  • Two fixed programs. run_proctor_check starts only your app's own PROCTOR, with an allowlisted environment. upgrade_path starts only Booster's read-only plan, never run. Both get an argument array and a timeout, and neither goes through a shell.

Adjoining sheets

The neighborhood

ADVISOR answers from a bundled copy of what these projects publish. Here is who lives where.

Here be roadmaps

Not yet surveyed

The edges of the map. None of these is built yet, so none of them gets a command here.

  • Coming soon

    Ask ADVISOR on the web

    A chat page. First we need to pick a model and decide what data it may see, and it will need its own accessibility review.

  • Coming soon

    Tools that write

    Things like starting an app with Sourdough or applying a Booster hop. When they come, they stay off until you set ADVISOR_READ_ONLY=0.

  • Coming soon

    HTTP transport

    stdio first. Remote needs a threat model before it needs a port.