The timestamp lands
Open the video at the minute quoted. If it points at silence, the alignment slipped and the whole report is suspect.
Felo YouTube Subtitling on Grok Bot
Grok Bot
Grok BotMedia watch is a routine job. Watching the video should not be.
A Bot that monitors a channel overnight is useful; a Bot that asks you to watch the upload is not. This pulls the captions as timestamped text, so the morning report says what was said, with the minute it was said in.
The request that saves the method
Save this as a skill called "Media watch". Pull the transcript, quote
anything that mentions us with its timestamp, and report nothing
on days with no upload.- Core unitsBot · skill · routine
- InputA channel or a video
- Comes back asTimestamped lines
- FitsA scheduled media watch
A Grok Bot owns a job, a conversation and a schedule, and a skill is the method it runs. Felo skills are the research half of that loop, saved once and reused by every routine.
- Core units
- Bot · skill · routine
- Skills come from
- A saved run, or Settings → Plugins → Yours
- Runs on
- iPhone, Android, and desktop
Timestamped lines, so the morning report can be checked in a minute
A quotation in a routine's output is read at breakfast and acted on at lunch. The timestamps are what let anyone verify a claim without watching the upload, which is the entire reason for the skill.
03:47Host
Before we start, a correction: the seat minimum we mentioned last week applies to the entry tier only.
12:29Guest
We tried the migration twice. The second time was manual, and that is the version that stuck.
18:03Host
So the automation was the problem, not the migration itself.
18:20Guest
For our data, yes. I would not generalise it to anyone else's.
41:15Guest
The pricing page changed again on Tuesday, which nobody has covered yet.
We tried the migration twice. The second time was manual, and that is the version that stuck.
Add it on Grok BotTeach the Bot your media-watch habit once
The method worth saving is not the fetch, it is your reporting rule: quote with timestamps, keep the caveats, and stay quiet on empty days.
- Run
Give one Bot a real job
Create a Bot that owns media watch for a defined set of channels, and say in the description that it reports only what was said, never what to do about it.
- Save
Ask it to record the method
Run one pass, ask for the transcript and the quotes with timestamps, then save the method with the reporting rule and the quiet-day rule attached.
- Enable
Turn the skill on where it is needed
Enable it on the monitoring Bot only. A transcript skill in a general chat agent invites quotations you did not ask for in threads that did not need them.
- Schedule
Attach a routine, then read the first run
Attach a routine that runs after the usual upload window, and read the first week. Confirm a day with no upload produces no message rather than an apology.
A media watch that stays readable in week twenty
The failure mode is a daily digest nobody opens. The rules below are what keep it short.
Runs when
A channel, a competitor, or a category where the useful information arrives in video.
- 01
Watch for uploads, not for keywords
Ask for a check per new upload. Keyword sweeps across a whole platform produce noise, and the routine gets muted by the second week.
- 02
Quote with timestamps, always
The minute is what makes the report actionable. A paraphrase of a forty-minute interview is an opinion about it.
- 03
Keep the hedge in the quotation
Speakers qualify themselves, and the qualification is often the most honest part. Cut it and the report is wrong in a way nobody notices.
- 04
Report nothing on empty days
A routine that must say something will invent a reason to speak. Silence on a quiet day is a feature you have to ask for.
Leaves behindA short report of what a new upload said about the things you care about, quoted with timestamps, and no message at all on days with nothing.
Three checks on a quoted claim
Everything a routine reports should be verifiable in under a minute, or the timestamps are decoration.
Names and numbers survived
Auto-captions mangle both. Check anything that will be repeated in a meeting against the audio.
The quiet days stayed quiet
Look back over a week of runs. If every day produced a message, the method is padding rather than reporting.
Two requests for the watch
Both assume the Bot is doing the watching, and both end with something you can verify rather than a summary you must trust.
Check this channel's new upload, pull the transcript, and tell me if it mentions our product — with the timestamp.What comes backEither a quoted mention with its minute, or a clear statement that the topic did not come up.
Quote anything from this episode about migration experience, with the sentence before it for context.What comes backShort, checkable quotations that include the qualification the speaker attached to them.
Three things a transcript routine cannot tell you
Video is a rich source with predictable weaknesses, and a routine hides weaknesses well.
Captions are not verbatim
Auto-generated captions are good enough to search and quote loosely. Anything that will be attributed to a named person should be checked against the audio.
No captions, no report
Recordings without subtitles give the routine nothing. It should say so rather than send an empty digest that looks like a quiet day.
A mention is not a verdict
Being named in a video is not the same as being criticised or endorsed. Ask for the sentence, not the tone.
What Grok Bot users ask about transcripts
Can the routine follow a whole channel?
Yes — that is the shape it is built for. Point the routine at the channels that matter, and have it check for new uploads rather than searching the platform.
How does it handle a video with no subtitles?
There is no text to fetch. The useful behaviour is a one-line note saying so, which is worth writing into the saved method.
Does the Bot need anything installed first?
Once: npm install -g felo-ai and felo config set FELO_API_KEY. The saved method calls that CLI, so the key lives in your user config rather than inside the skill.
Should the full transcript be kept?
Keep it with the run if you might need to quote more later, and keep the quotations in the report itself. Both, if the channel is important.
Can it translate a foreign-language recording?
You get whatever captions the video carries, and the report can be written in your working language with the original quote beside it.
Every Felo skill for Grok Bot
Five skills written for Grok Bot. They follow the same pattern on this platform, so you can combine them in one session without re-explaining your setup.
Felo Search
Felo Slides
Felo X Search
Felo SuperAgentOne API key. A full agent toolkit.
Start with a skill, then compose the exact models, live data, knowledge, and output tools your Grok Bot workflow needs — the same key every Felo skill already uses.
Models
Use cost-efficient LLM access when your agent needs custom reasoning or generation.
- LLM API
- Chat completions
- Model access
Live information
Give agents current web and social signals instead of relying on stale context.
- Web Search
- X Search
- Web Fetch
Knowledge and output
Ground work in reusable knowledge, then turn it into shareable deliverables.
- LiveDoc
- PPT generation
- Mindmaps
Creative tools
Extend the workflow from research to visual and office-ready results.
- Image generation
- Page tools
- Office outputs