Getting started.
Nothing to install and nothing to keep up to date. Connect your AI model once; from then on it fetches any skill it needs, whole and current.
1. Connect to your AI model
Once per app, two ways. Either way you sign in with your Google account when asked, and access follows your account, not your machine: being on the team roster is what lets you in.
Let your agent set it up
Copy the prompt and paste it into an agent that runs commands: Claude Code, Codex, Gemini CLI or Cursor. It adds the connector, tells you when to sign in, and checks the connection.
Or do it yourself
Add the connector by its address in the app’s own settings, then sign in.
https://skills.ascendsecrets.com/mcpClaude Code
- In a terminal:
claude mcp add --transport http ascend-skills https://skills.ascendsecrets.com/mcp - Inside Claude Code, run
/mcp, pick ascend-skills and sign in.
Claude Desktop and the Claude web app
- Settings, then Connectors, then add a custom connector.
- Paste the address, connect, and sign in when asked.
ChatGPT
- Settings, then Connectors. If there is no way to create one, turn on developer mode under Advanced first.
- Create a connector with the address, then sign in when asked.
GPTwork
- Add a connector by its address.
- Sign in when asked.
Whichever app you use, the same two screens follow: Google asks which account, then the library’s own page asks whether that app may read skills as you. Allow it only if you started it yourself, a moment ago. If your account is not on the roster, that page says so and nothing is granted; ask a platform admin to add you.
2. Use a skill
20Ask your AI for help the way you normally would. It finds the right skill in the library, reads it exactly as it would read an installed one, and does the work with you. Every use fetches the current version, so an improvement reaches you the next time you ask. To point it at one skill, open that skill’s page here and press Copy prompt: the line it copies names the skill, and your AI asks you for whatever the skill needs. This site shows what each skill does; the skill itself only ever reaches your AI through the connector.
If your AI answers “no release is published”, the library has not been published to the connector yet. Ask a platform admin.
Configuring a skill
Skills never ship client data. Anything client-specific, such as cost-per-result targets, ad account ids or sheet ids, comes from a config file in your project root, which is gitignored:
.ascend/
├── skills.json status file location, Slack pulse settings
├── media-buying.json per-client targets, account ids, tracker sheets
├── clients.json Drive folder ids
└── notion.json database and data-source idsEach skill documents the shape it expects. If the config is missing, a skill will ask rather than assume a default: a wrong target silently turns a kill decision into a keep, which is worse than an error.
Improving a skill
You never need git. A new skill, or a better version of an existing one, is uploaded on the internal contribution screen, checked automatically, and listed once a platform admin has approved it. Until that screen exists, send the skill to a platform admin, who opens the pull request for you.
Whoever does the editing works to CLAUDE.md in the repository: it carries the Portability Checklist, and every item on it is there because it caught a real bug. Each change bumps the skill’s version and adds a CHANGELOG.md entry.
