vibe/code/agent
Shelf Pass $1.95/mo Free during preview Shelf

For people who build with AI coding agents

Brief the agent.
Check the work.

Write the one file every coding agent reads first, brief each task so “done” means something, and run 24 checks before agent-written code goes live. No AI here, and nothing you type is sent to an AI company or any other outside service.

Write your AGENTS.md Run the ship check

The tools are free. Shelf Pass $1.95/month or $19.50/year saves projects and versions · Free during preview

AGENTS.md20 lines
# Corner Bakery ordersOnline cake orders for one bakery.## Commands| Task | Command || --- | --- || Test | `php artisan test` |## Rules### Safety- Never write secrets into code.- Do not run commands that deploy.## Done means- Tests pass; changed files listed.
  • Codexreads AGENTS.md
  • Claude CodeCLAUDE.md: @AGENTS.md
  • GitHub Copilotreads AGENTS.md
  • Cursorreads AGENTS.md
  • Gemini CLIcontext.fileName
  • Aiderread: AGENTS.md
One rulebook, wired to six agents the way each maker documents it. See the field guide.

Three files. One afternoon.

Agents follow what is written down. Write it down once.

  1. 01

    Rulebook

    Name the project, its commands and folders, pick rule groups (safety, scope, tests, git, dependencies, data, style) and say what “done” means. Out comes one AGENTS.md plus the small files that point Claude Code, Gemini CLI and Aider at it.

    Write the rulebook
  2. 02

    Work order

    Brief one task: goal, where to look, “done when” criteria, what is out of scope and when to stop and ask. Brief Check flags every criterion nobody could check, like “make it better”.

    Brief a task
  3. 03

    Ship check

    24 yes-or-no points in six areas, before the code goes live. Each gap comes with why it matters, how to fix it, and a prompt you can paste into your own agent to look for it.

    Check before shipping

Brief Check, on a real example

“Better” is not a test.

Brief Check reads each “done when” line and looks for something you could see or measure: a number, a status, a path, quoted text, or a verb like returns or shows. These are its actual results for the example brief.

4of 5

criteria are checkable

Open this example
  • checkable

    GET /forgot shows a form with one email field

    Checkable: it names a path or file.

  • checkable

    Submitting a known email sends one message containing a link that expires in 60 minutes

    Checkable: it names a number.

  • checkable

    Submitting an unknown email shows the same "Check your inbox" message (no account leak)

    Checkable: it names quoted text.

  • checkable

    The reset link works once; a second use shows "This link has expired"

    Checkable: it names quoted text.

  • vague

    The login page looks better

    "Better" cannot be checked. Say what you will see: a number, a message, a status code, a file, a test that passes.

Field guide

Which file does each agent read?

Full guide with sources
AgentReadsTo use one AGENTS.md
CodexOpenAIAGENTS.md (or AGENTS.override.md, which wins in the same folder)Nothing to add: AGENTS.md is read as it is.
Claude CodeAnthropicCLAUDE.md, .claude/CLAUDE.md and CLAUDE.local.md; AGENTS.md when there is no CLAUDE.mdA CLAUDE.md holding the line @AGENTS.md imports the rulebook.
GitHub CopilotGitHub.github/copilot-instructions.md, .github/instructions/*.instructions.md, AGENTS.md, or a single CLAUDE.md or GEMINI.md at the rootNothing to add: AGENTS.md is read as it is.
CursorCursor.cursor/rules/*.mdc, and AGENTS.mdNothing to add: AGENTS.md is read as it is.
Gemini CLIGoogleGEMINI.md by default.gemini/settings.json with context.fileName listing AGENTS.md (and GEMINI.md).
AiderAiderAny conventions file you load: /read AGENTS.md, aider --read AGENTS.md, or read: in .aider.conf.yml.aider.conf.yml with the line read: AGENTS.md loads it in every session.

Checked against each maker's own documentation on October 11, 2026. Tools change; the field guide links every page.

Ship check

Before it goes live: six areas, 24 questions.

Answer yes, no or not sure. The sheet says how many points match this list, never that the code is “secure”, and turns every gap into a fix and an audit prompt.

  • Secrets

    No key, token or password is in the repository or its history.

    4 points

  • Access

    Every page and API route that shows or changes private data checks, on the server, who is asking.

    4 points

  • Input and data

    Every database query uses parameter binding or the ORM; no SQL is built from user input.

    4 points

  • Dependencies

    Every package the agent added exists on the official registry and is the one you meant.

    4 points

  • Configuration

    Debug mode and detailed error output are off in production.

    4 points

  • Change control

    The code is in version control and you can get back to the last working version.

    4 points
Start the ship check

Pricing

The tools are free. The Shelf Pass keeps your work.

Free during preview Checkout is not open yet. Every Pass feature is free for everyone until it is, and nothing is charged. The prices below are the prices it will have.

free
$0every tool, no sign-up
shelf-pass
$1.95a month
or yearly
$19.50a year, two months free
Start writing Pricing details
free..shelf-pass.diffwhat the Pass adds
 Both plans: Rulebook: AGENTS.md + wiring files, copy or download each  Both plans: Work order + Brief Check, unlimited  Both plans: Ship check sheet + audit prompt  Both plans: Field guide -Free plan only: Shelf: 1 project, newest 3 rulebook versions, 3 work orders, 1 ship check +Shelf Pass: Shelf: 25 projects, 30 rulebook versions each +Shelf Pass: Line-by-line compare between versions +Shelf Pass: 100 work orders and 12 ship checks per project +Shelf Pass: The .zip starter kit: every file at its right path +Shelf Pass: 7 stack packs (Laravel, Django, Rails, Next.js, CodeIgniter 4, Go, Python)

Free during preview No checkout yet: every + line is yours now at no charge.

Questions

Does Vibe Code Agent use AI or write code?

No. Every result comes from fixed rules, and the work-order page lists every word it looks for. The site writes instruction files and briefs; your own agent writes the code.

Is anything I type sent to an AI company?

No. The tools run on this server and answer in the page. Nothing is sent to Codex, Claude, Copilot, Cursor, Gemini, Aider or anyone else. Unless you save to a Shelf, what you type is not stored; the web server keeps standard access logs (see Privacy).

What if I paste a password or API key by mistake?

Secret Guard looks for key and password shapes (cloud keys, tokens, private keys, passwords in URLs) and replaces them with [REMOVED: …] before any file is written. A Shelf refuses to save them.

Does passing the ship check make my app safe?

No list can promise that. The sheet tells you which of 24 common, well-documented points your answers match, and what to do about the rest.

Is this the Vibecode app?

No. Vibe Code Agent is an independent WeldonPC.com site and is not connected to Vibecode (vibecodeapp.com), Mistral Vibe, Salesforce Agentforce Vibes, or any company whose agent it names.