What they did
The poster runs a small YouTube character channel, Cynic Stick, almost entirely from n8n. Much of the setup was built by Claude through the n8n MCP server, with the poster directing. About 25 workflows run daily and weekly: scheduled trend searches feed a research step (with Ollama as a fallback), one workflow writes and renders the daily short, and the weekly long video is split into script, scene prompts, renders, compile and review. A single registry workflow tells every other one which models to use. Separate workflows publish to YouTube, X, Instagram and Facebook on a schedule, while TikTok uploads go to drafts and the caption arrives on Telegram for a manual post. A watchdog checks each build 30 minutes before publish time, and a daily job files shorts into playlists. A newer set of 12 workflows runs a live stream: viewers type a command and an idea in YouTube chat, a poller picks it up, several OpenRouter judges screen it, a model turns it into a scene for an ASCII-art world, and the poster gets a Telegram veto button before it airs.
How it went
The poster describes the pipeline as running day to day, with the watchdog only messaging when a build did not start or finish, so silence means it worked. TikTok still needs one manual tap. The live stream is brand new and the audience is tiny, and the poster gives no view, cost or time-saved numbers. They are asking the community whether an hourly moderation test is enough.
Worth knowing
The moderation is built to fail closed: anything unclear is rejected, an hourly canary sends four fixed lines through the judges and locks the stream if a bad one passes, repeated human vetoes also lock it, and there is a kill switch. Raw chat text is deleted after 7 days and names after 30. Polling live chat has to be balanced against the YouTube API quota.