tokensagents .COM
Install

Installed in about ten minutes.

Pick your tool and open its folder. In Claude the skills install once and fire on their own. In Cursor, Codex, and other coding agents a router file points the agent at the same rules. In any chat, you paste a prompt. Three paths, same rules, about ten minutes.

Install, by tool

Claude Code and claude.ai, skills that fire on their own

  1. Unzip the two skill files.

    Open the 1-claude folder. Each skill ships as a zip of the skill folder: ada-visual.zip and ada-aria.zip.

  2. Put each skill folder in your skills directory.

    In Claude Code, personal skills live in ~/.claude/skills/ and load in every project; project skills live in .claude/skills/ and ship with one repo. Either works. 

    mkdir -p ~/.claude/skills
unzip ada-visual.zip -d ~/.claude/skills/
unzip ada-aria.zip -d ~/.claude/skills/

    You should end up with ~/.claude/skills/ada-visual/SKILL.md and ~/.claude/skills/ada-aria/SKILL.md. On claude.ai or the Claude desktop app, skip the terminal: go to Customize, then Skills, then Create skill, then Upload a skill, and upload each zip. Enable code execution once under Settings, then Capabilities.

  3. Point one at your code.

    Name a file or a pull request, or just ask for a review. The skill fires on matching work on its own, and you can always call it by name. Installs do not sync between surfaces, so install on each one you use. 

    Run an ADA check on src/components/Dialog.tsx
Do a full accessibility review of this component at AA
Regulatory check only on this page

Cursor, Codex, and other coding agents, one router file

Open the 2-cursor-codex-and-other-agents folder. It carries an AGENTS.md router and the same rule modules. Copy them into the repo your agent works in, with AGENTS.md at the repo root. OpenAI Codex, GitHub Copilot's coding agent, Cursor, Windsurf, and Zed read AGENTS.md natively, nothing to configure. Two notes: Google Gemini CLI reads GEMINI.md by default, so add AGENTS.md in settings.json with "context": {"fileName": ["AGENTS.md", "GEMINI.md"]}. And Zed prefers an existing .rules or .cursorrules file over AGENTS.md, so remove those if you want Zed reading the pack.

One difference from Claude: firing on their own is a Claude feature. In a coding agent you include the rules with the work rather than having them trigger automatically.

ChatGPT, or any chat, nothing installed

Open the 3-chatgpt-and-any-chat folder. Paste prompt 02, the full component audit, on a component and work the findings top down, or paste prompt 00 to route a whole review. Prompts 01 through 06 each carry a self-contained version that holds its own numbers, so it works cold in any chat with nothing else loaded. Your code goes to that provider, so use this where that is allowed.

On the roadmap: an optional MCP server, run locally in your own environment, that serves the rules to any MCP-capable agent so a non-Claude tool can query them like an API instead of loading files by hand. It stays local, and nothing about the security story below changes.

What IT will ask

  • The files are static and plain text. Markdown files, one folder per format. Your team can read every rule before installing it.
  • They run inside your own environment. The pack has no server and makes no network calls of its own.
  • No telemetry, no vendor access. We never see your code and get no telemetry. The chat prompts, like any chat, send your code to that provider; the Claude and coding-agent formats stay where your agent runs.
  • License enforcement lives with the merchant of record. Polar handles keys and renewal. The files do not phone home to check a license.
  • Verify your download. Run shasum -a 256 on the zip and match the SHA-256 we publish. Individual pack eccc4334947172ac51cc02ac7950af08cf4291362cc1c348dfdc737411badf6f Team pack 0a3dc1d4c3f5c4dee07f85decf036b6b5396233b67b3140bac7cb06701e36092

For a managed Claude Code fleet, the pack can be pinned and allowlisted so only an approved version runs. Enterprise licenses include the exact configuration snippet for your admin.

Requirements

  • An AI agent that reads Markdown. Claude runs the auto-triggering skills; Cursor, Codex, and other coding agents use the router; any chat runs the prompts.
  • For repo fixes, that agent on the developer machine, Claude Code or your editor's agent.
  • Platforms the pack reviews: web, iOS, Android, desktop, and enterprise internal tools. Any stack, any framework. If it renders a UI, the rules apply.

Ready when your repo is.

Get the pack