Skip to content
Get packs

Guide · Agent skill

How to use the Google Workspace Memory agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The google-workspace-memory skill lets an AI agent remember your Google Calendar, recent Gmail and one Drive folder in a single cognee memory, then answer questions across all three. It is aimed at professionals who want one briefing that combines meetings, email and documents, and at developers testing the cognee Google Workspace cookbook. After this guide you will know what the skill does, how to install it, how to trigger it with plain requests, and what it needs before it runs.

Google Workspace Memorytopoteretes/cognee · Apache-2.0

Quick start

  1. 1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or the project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills instead.
  2. 2Prepare the cognee repository. Check out the cognee repository and make sure Python with uv is installed. Run all commands from the repo root.
  3. 3Set your LLM key. Add LLM_API_KEY for your LLM provider. Without Google credentials, the skill runs on sample data and only this key is needed.
  4. 4Add Google credentials for real data. For your own data, place credentials.json with Calendar, Gmail and Drive read-only consent in the cookbook folder and set GOOGLE_DRIVE_FOLDER_ID, or use --no-drive.
  5. 5Run the setup check. Run the --check command. Fix any line marked [setup] MISSING before going further.
  6. 6Ask your question. Make a plain request such as preparing for next week's client meetings. Approve the browser consent prompt for each app on the first run.

What is an agent skill and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent when to use a capability and how to run it. This skill runs the Google Workspace memory cookbook from the cognee repository. It remembers your primary calendar, newest inbox emails and one Drive folder in the cognee dataset google_workspace_memory, then answers one question across them.

By default it prepares you for client meetings in the coming week, such as which orders are late or at risk, why, and what to say. Google access is read-only, so nothing is created, changed or sent.

  • Reads the primary Google Calendar, newest Gmail inbox emails and one Drive folder.
  • Stores them in a cognee memory and answers a question across all three sources.
  • Runs locally, and reads Google data read-only.
  • Without your Google setup, it runs on a built-in sample about a fictional ceramics business.

Install the skill in Claude Code or the Claude apps

Download the ZIP from this page. You can install it for a single user or for one project.

For a personal install, unzip it into ~/.claude/skills/. For a project install, unzip it into the project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/
  • Project: .claude/skills/ inside your project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What you need before the first run

The skill needs a checkout of the cognee repository, because it runs the cookbook in examples/cookbooks/google_workspace_memory/. Python with uv must be installed, and the commands run from the repo root.

An LLM_API_KEY for an LLM provider is required. For real data you also need a Google Cloud credentials.json with Calendar, Gmail and Drive read-only consent, plus GOOGLE_DRIVE_FOLDER_ID unless you skip Drive with --no-drive. The cookbook's README has the full 'What it needs' table.

  • Python with uv installed
  • cognee repository checkout
  • LLM_API_KEY for an LLM provider
  • Google Cloud credentials.json with Calendar, Gmail and Drive read-only consent (real data only)
  • GOOGLE_DRIVE_FOLDER_ID (unless --no-drive is used)

Trigger the skill with natural requests

You do not need special syntax. The skill is meant to be used when someone asks to connect, ingest or sync their Google Workspace, or asks a question that needs calendar, email and Drive together.

Ask in plain language, for example about the coming week's client meetings or a question that spans your inbox and files. The agent then runs the checks and the cookbook script itself.

  • "Connect my Google Workspace and prepare me for next week's client meetings."
  • "Sync my calendar, email and Drive, then tell me which orders are late or at risk and why."
  • "Using my calendar and emails, what should I say to the client on Thursday's call?"
  • "Try the Google Workspace memory on sample data first."

Run the skill step by step

The agent first runs a check. If it passes, it runs the main script, which ingests the sources and answers your question. The first run opens the browser once per app for consent, so approve Calendar, Gmail and Drive when prompted.

Without credentials.json, the run uses the sample workspace and says so. You can add --sample to use the sample even when your account is set up. Follow-up questions on the same data can use the ask script without ingesting again.

  • Check setup first; exit 2 means a named item is missing.
  • Use --no-calendar, --no-gmail or --no-drive to skip any app.
  • Use --ask "..." to pass your question.
  • Follow-up questions use scripts/ask.py and do not ingest again.
uv run python examples/cookbooks/google_workspace_memory/google_workspace_memory.py --check

Safety notes: scripts, network and your data

This skill runs scripts and uses the network. Ingestion reads your real calendar, mailbox and Drive files in full, and every run uses LLM credits. The first calendar ingest can take ten minutes or more.

The agent should not print the contents of credentials.json, the token files or .env. A sample run clears only the sample dataset before ingesting. Clearing your own dataset with --clear forgets everything it holds, so use it only when you want to start over.

  • Google access is read-only.
  • Do not share credentials.json, *_token.json or .env.
  • Real-data ingestion reads full content, so consider the sample first.
  • To remove stored data, run the cognee-cli forget commands for each dataset.
uv run cognee-cli forget --dataset google_workspace_memory

What people use it for

  • Preparing for the coming week's client meetings with calendar, email and Drive context together.
  • Finding which client orders are late or at risk, and why, across mail and shared documents.
  • Testing the cognee memory cookbook with Google data before using real accounts.
  • Running a safe demo on the built-in fictional ceramics sample.

Frequently asked questions

How do I install the google-workspace-memory skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills.

What do I need to run the Google Workspace memory skill?

You need Python with uv, a cognee repository checkout and an LLM_API_KEY. For real Google data you also need credentials.json with read-only Calendar, Gmail and Drive consent, plus GOOGLE_DRIVE_FOLDER_ID unless you use --no-drive.

Is the Google Workspace memory skill free?

The source does not state a price or licence terms for the skill beyond the repository licence, which is Apache-2.0. Every run uses LLM credits from your LLM provider, so check that provider's pricing.

Can I try it without connecting my Google account?

Yes. Without credentials.json in the cookbook folder, the skill runs on a built-in sample about a fictional ceramics business and only LLM_API_KEY is needed. The answer is about sample data, not your real files.

Does the skill read or change my Google data?

Google access is read-only, so nothing is created, changed or sent. However, ingestion reads your calendar, mailbox and Drive files in full, so only use it if you are comfortable with that.

Why is the first run so slow?

The first run cognifies your whole primary calendar, which can take ten minutes or more, and every run uses LLM credits. Run it only when you need it and tell the agent to start it deliberately.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you want one briefing that combines your calendar, recent email and a Drive folder, and you are comfortable with an AI pipeline reading those in full. Developers evaluating the cognee cookbook will also find the sample run useful. If you need instant answers, or do not want your real mail and files processed, pick a simpler tool or stay on sample data.

Download the skill