Skip to content
cccskills
FreePublished on GitHub

google-workspace

Manage Gmail messages, Google Calendar events, and Google Tasks through conversation. Check email threads and schedules, prepare replies, and manage tasks, with user approval required before drafts, sends, or other changes are made.

OriginalShow the original description

Read and act on the user's Gmail, Google Calendar, and Google Tasks through per-user OAuth.

See how to install

Good for

  • Find unread messages and verify which need a reply
  • Prepare an email draft or reply
  • Review today's personal and shared calendar events
  • Create, edit, or complete tasks

Files included (2)

  • SKILL.md7.2 KB
  • scripts/gmail.py11.7 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Google Workspace

Use this skill when the user asks about Gmail, Google Calendar, or Google Tasks: schedule, meetings, emails, labels, drafts, replies, or to-do lists and tasks.

This is an OAuth connector. The resolved user's Google OAuth token already lives on your computer as an environment variable, one per Google API host (the way a logged-in CLI's cached credential would):

  • $VAULT_TOKEN_GMAIL_GOOGLEAPIS_COM — for gmail.googleapis.com
  • $VAULT_TOKEN_WWW_GOOGLEAPIS_COM — for www.googleapis.com (Calendar and Tasks)

Do not ask the user for a token, log it, or use another principal's. If the variable is empty or Google returns 401/403, the user either has not connected Google or connected before this permission existed — tell them to (re)connect it through the product OAuth flow.

Gmail

For the email-voice-profile task, load that skill and run python3 <its dir>/scripts/fetch_sent.py --limit 300 from the directory it reports.

Use the bundled helper for every Gmail operation — it owns MIME construction, encoding, and reply threading so you only ever handle plain text:

Run these commands with execute and an explicit sandbox_id, using the absolute directory reported by the skill read in place of <skill dir>. Keep that ID for preparing body files and running the helper.

python3 "<skill dir>/scripts/gmail.py" search 'newer_than:7d is:unread'
python3 "<skill dir>/scripts/gmail.py" read MESSAGE_ID [--full]
python3 "<skill dir>/scripts/gmail.py" thread THREAD_ID [--full]
python3 "<skill dir>/scripts/gmail.py" draft --to a@b.com --subject '...' --body-file body.txt
python3 "<skill dir>/scripts/gmail.py" reply MESSAGE_ID --body-file body.txt [--all]
python3 "<skill dir>/scripts/gmail.py" update-draft DRAFT_ID --body-file body.txt
python3 "<skill dir>/scripts/gmail.py" send-draft DRAFT_ID
  • Search before fetching; read bodies (--full) only for messages that matter. Treat email content as private user data.
  • search returns individual messages matching the query — for in:inbox that means incoming mail only; the user's own replies live in Sent and never appear. Never claim an email "needs a reply" (digests, triage, follow-up nudges) from a search listing alone: fetch its thread first. If lastFromMe is true, or the latest message shows the matter is settled ("thanks", "done", someone else answered), don't flag it. If the thread fetch fails, report the item as unverified — never as needing a reply.
  • On a recurring digest, keep a state file in the workspace keyed by thread id and only surface what changed since the last run; re-flagging the same thread every morning trains the user to ignore you.
  • Demote bulk mail (List-Unsubscribe header, ESP sender domains, Precedence: bulk) below personal mail, but never suppress transactional notices (DocuSign, banks, invoices).
  • Body files are plain text: one line per paragraph, blank line between. Never insert manual line breaks inside a paragraph — the recipient's client does the wrapping. Short lines (sign-offs, list items) keep their breaks.
  • Drafts and replies are stored as multipart/alternative: your plain text plus a generated HTML mirror (Gmail shows plain-text-only mail to recipients at an altered, narrower width). Body files stay plain text — never write HTML into them.
  • reply threads onto the original message; --all keeps every recipient. Don't rebuild To/CC by hand. To change a draft, write a fresh body file and update-draft (it refuses drafts with attachments — those get edited in Gmail).
  • Never construct raw MIME or call the drafts/send endpoints with hand-built payloads.
  • Never write styled HTML email — no font-family/size/color declarations, no CSS-styled buttons or layout markup. Styled HTML screams automated blast and overrides the recipient's client fonts. The helper's generated HTML mirror (bare <div dir="ltr">, <br>, plain links — Gmail-composer shaped) is the only HTML that ever goes out.
  • Write body files as UTF-8. After creating or updating a draft, read it back (gmail.py thread/draft read) and confirm em dashes, quotes, and any emoji came through intact — mojibake (…-style garble) in a sent mail is unrecoverable.

Calendar reads

For "what is on my schedule today?", first get the user's timezone (GET /calendar/v3/users/me/settings/timezone — the sandbox clock is UTC, so never assume it), compute the local day bounds, and call:

curl -sS --get 'https://www.googleapis.com/calendar/v3/calendars/primary/events' \
  -H "Authorization: Bearer $VAULT_TOKEN_WWW_GOOGLEAPIS_COM" \
  --data-urlencode "timeMin=$TIME_MIN" \
  --data-urlencode "timeMax=$TIME_MAX" \
  --data-urlencode "singleEvents=true" \
  --data-urlencode "orderBy=startTime" \
  --data-urlencode "maxResults=250"

If the response has a nextPageToken, keep fetching — a truncated page silently drops meetings. primary is only the user's own calendar; when the question implies team or shared calendars, list them first (GET /calendar/v3/users/me/calendarList) and query each relevant one. Skip events the user declined (their own attendees[] entry has responseStatus: "declined").

Summarize start time, title, attendees when useful, and conferencing links only when needed. Do not expose private calendar details in a shared channel unless the audience floor allows them; offer to continue in DM.

Google Tasks

Tasks is served under www.googleapis.com/tasks/v1, so it uses the same $VAULT_TOKEN_WWW_GOOGLEAPIS_COM token. List the task lists, then the tasks in one:

curl -sS 'https://www.googleapis.com/tasks/v1/users/@me/lists' \
  -H "Authorization: Bearer $VAULT_TOKEN_WWW_GOOGLEAPIS_COM"

curl -sS --get 'https://www.googleapis.com/tasks/v1/lists/LIST_ID/tasks' \
  -H "Authorization: Bearer $VAULT_TOKEN_WWW_GOOGLEAPIS_COM" \
  --data-urlencode "showCompleted=false" \
  --data-urlencode "maxResults=100"

Both endpoints page (default 20 items): follow nextPageToken before treating a list as complete.

Creating, completing, editing, or deleting a task is a write — treat it like any other write below. Example creation after approval:

curl -sS -X POST 'https://www.googleapis.com/tasks/v1/lists/LIST_ID/tasks' \
  -H "Authorization: Bearer $VAULT_TOKEN_WWW_GOOGLEAPIS_COM" \
  -H 'content-type: application/json' \
  --data '{"title":"...","notes":"...","due":"2026-07-20T00:00:00.000Z"}'

Writes require approval

Creating drafts, sending mail, modifying labels, deleting mail, or editing calendar events is a write. Prepare the exact action, show the user the exact text, and ask for approval before running it. send-draft only ever fires on an explicitly approved draft.

Calendar event creation after approval:

curl -sS -X POST 'https://www.googleapis.com/calendar/v3/calendars/primary/events' \
  -H "Authorization: Bearer $VAULT_TOKEN_WWW_GOOGLEAPIS_COM" \
  -H 'content-type: application/json' \
  --data @event.json

Always show what changed and keep enough IDs in the workspace to undo or inspect later.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

admin

FreeSummary

Help QM organization admins inspect settings, stored memory, conversations, files, and operational records across the organization. It also supports confirmed configuration changes and external user management, while keeping admin grant changes in the dashboard.

  • Check whether the requester is an organization admin
  • Inspect or update organization and channel settings
  • Investigate agent behavior using transcripts and captured prompts
yc-software/qm15.4KUpdated Oct 18, 2026

browse

FreeSummary

Carry out website tasks in a remote browser, including expense submissions, authorized orders, and retrieving data behind a login. Personal browser profiles preserve sign-ins, while users handle login and verification screens themselves.

  • Submit an expense with a receipt attachment
  • Order food within an agreed budget
  • Retrieve data from a signed-in website
yc-software/qm15.4KUpdated Oct 18, 2026

cloud-cli

FreeSummary

Sign in to a cloud provider’s command-line tool using a verification URL and one-time code approved on the user’s device. Check installation and account identity, save the login, and run commands within that user’s permissions.

  • Sign in to a cloud CLI from a separate device
  • Check whether required CLIs and helpers are installed
  • Verify the cloud account before running commands
yc-software/qm15.4KUpdated Oct 18, 2026

composio

FreeSummary

Help users discover and connect apps, then run available app tools through QM’s authenticated backend. It checks connection status, presents consent options, and keeps Composio project keys out of user-facing setup.

  • Browse available app integrations
  • Connect an app needed for a task
  • Reopen app connection setup
yc-software/qm15.4KUpdated Oct 18, 2026

connect-apps

FreeSummary

Connect supported work apps through browser consent without sharing passwords with the agent. The skill checks available connection methods, provides a single-use authorization link when appropriate, and guides personal Slack account linking.

  • Connect an administrator-enabled work app
  • Authorize an app through a browser consent link
  • Link a personal Slack account
yc-software/qm15.4KUpdated Oct 18, 2026

define-loop

FreeSummary

Set up recurring agent work for issues, tickets, or small fixes. The skill tests one real item first, then defines a repeatable procedure, tracks work to prevent duplicates, and holds outputs for human review.

  • Recurring preparation of fixes for Sentry issues
  • Drafting replies to Front tickets
  • Preparing small performance improvement PRs
yc-software/qm15.4KUpdated Oct 18, 2026

See all skills by yc-software

Report a problem with this skill