vellum-ai/meet-join

Latest recorded source activity: July 1, 2026 at 21:09

SkillsMP catalog refreshed: November 9, 2026 at 14:22

skills collected: 1

GitHub stars: 2
GitHub forks: 1

Install with an AI assistant

Copy this prompt into the AI assistant you're using.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/vellum-ai/meet-join.

Repository README

Repository README source: README.md

Read the source README

Expand full README

# meet-join Google Meet join + transcription + voice/chat participation skill. The skill entrypoint lives in [`skills/meet-join/SKILL.md`](./skills/meet-join/SKILL.md); skill-internal architecture and the isolation rule are covered in [`AGENTS.md`](./AGENTS.md). ## Docs - [Contributing / testing a dev build](./CONTRIBUTING.md) - how to install and test a development version of this plugin from a GitHub URL. - [Live verification runbook](./docs/LIVE-VERIFICATION.md) — manual smoke tests for multi-party scrapers, streaming STT, barge-in, and consent-triggered auto-leave. Run these against a real Meet whenever you touch the bot, extension, audio ingest, or consent monitor. ## Performance on arm64 hosts The bot image is built and run with `--platform linux/amd64` because Meet's BotGuard accepts a plain-subprocess Chromium launch, and Chromium for Linux/amd64 is the platform with the richest working set of base-image packages (Debian ships a current `chromium` for amd64 but lags for arm64, and `google-chrome-stable` dropped `--load-extension` support that the extension-based join architecture depends on). On an arm64 host (Apple Silicon Mac, Graviton, Ampere) the image therefore runs under qemu user-mode emulation, which carries a meaningful CPU tax on bot-boot steps like Bun startup and Chromium V8 initialization. To quantify the overhead per image build and spot regressions, use [`bot/scripts/bench-join-latency.sh`](./bot/scripts/bench-join-latency.sh) (docs: [`bench-join-latency.md`](./bot/scripts/bench-join-latency.md)). The script runs N container launches against a test Meet URL and reports per-run CSV plus mean / median / p95 for two intervals: - `booted_delta`: container start → `meet-bot booted` (PulseAudio up, pre-Chromium). Measures base container-startup + qemu interpreter warmup. - `ready_delta`: container start → `meet-bot ready (meetingId=…)` (extension loaded, join command dispatched). Adds Chromium launch and Meet prejoin DOM time; the closest stdout proxy we have for "about to click Join". ### Baseline numbers _To be filled in after a live smoke-test run on the reference dev hardware (Apple Silicon M-series under macOS Docker Desktop with qemu-user-static). Expect order-of-magnitude overhead on `booted_delta` vs. a native amd64 host and 3–5× on `ready_delta`._ ``` # image=vellum-meet-bot:<tag> iterations=<n> meet_url=<test-room> # booted_delta (start → meet-bot booted): n=<n> mean=<m>ms median=<m>ms p95=<m>ms # ready_delta (start → meet-bot ready): n=<n> mean=<m>ms median=<m>ms p95=<m>ms ``` If the arm64 cost becomes a blocker (latency visible to the user, CI runtime blowup, flaky smoke tests), the path forward is a native arm64 image: Chromium from an alternate source (e.g. `chromium-browser` via `snap` is off the table in containers, but `playwright`-style chromium builds or the `browser-use` community images ship arm64 variants). That work is deliberately deferred until we have baseline numbers showing it's worth the complexity.

Loading the repository README…

Skills in this repository

Software Developers

1 occupation category · 100% classified

skill

occupation

description

updated

meet-join

Software Developers

Join a Google Meet call to take notes; only when the user explicitly asks.

2026-07-01

Showing 1 of 1 collected skill.