Periodic email check for a TentClaw CLAW, not a channel connector: cron runs fetch --peek-due every 5 minutes; every EMAIL_CHECK_MINUTES (.env, default 30) it checks and rings the channel's chime (content-free) only for new mail no chime announced yet, plus a daily chime if the check fails in any way; fetch keeps its own log (--log), so the lines work on a fresh install. The agent runs fetch, which shows only new mail since a UIDVALIDITY-aware cursor (first run and resets start from now), read-only, framed as untrusted outside text (<channel neutralised, fields escaped, HTML converted linearly and capped), text a person would not see (hidden styles, background-colored text, readable base64, a text part the HTML does not show) removed and reported on a hidden: line, spam skipped; the cursor moves only after the output was written. reply <uid> <file> answers one message over SMTP, explicitly: one address by default, a differing Reply-To only with --to-reply-to, authenticated, and it refuses credentials in the text, automatic, bulk and list mail, bounces, spam, 3-reply loops between CLAWs and recipients outside REPLY_ALLOW. Settings from .env only. Python 3 standard library, no dependencies.

Chrome on a virtual display in the tent, for the CLAW to drive over CDP (browser open, eval, text, shot) and for a person to sign in through noVNC, which ./viewer switches on for one sign-in, behind a fresh token per session, and off again. CDP has no TCP port: Chrome's debugging pipe behind a Unix socket only the tent user opens; the screen is on a Unix socket too. A tent service; install adds Chrome's libraries, Xvfb, x11vnc and noVNC.

Google Workspace CRUD via the googleworkspace/cli (gws) binary, pinned to a skill-local config directory so credentials and binary survive container rebuilds. Use when creating, reading, writing, or sharing Google Docs, Slides, Sheets, Drive files, or when sending Gmail / managing Calendar.

Fetch and extract content from URLs. Use when reading a webpage, extracting article text, fetching URL content, or retrieving tweet/X post data.