Get the transcript of this talk and add the three most useful points to ~/Documents/reading-notes.md with timestamps.What comes backYour notes file updated with anchored quotations, in the format you already use.
QoderWork runs on your machine and edits local docx and pptx files. A Felo skill supplies the half a local file cannot: current sources, transcripts, and assets that land beside your document.
The install is a single request. After that, the useful habit is to name both the video and the destination file in the same sentence, so the output lands where you keep notes.
Please download https://github.com/Felo-Inc/felo-skills/tree/main/felo-youtube-subtitling and put it in ~/.qoderwork/skills/, then confirm the skill is loaded.
Cloned the repository, placed felo-youtube-subtitling in ~/.qoderwork/skills/, and confirmed it is loaded. It reads captions and returns timestamped text.
Get the transcript of this talk and add the three points that matter to ~/Documents/reading-notes.md, with the timestamps.
Added three quoted points to reading-notes.md, each with its timestamp. The rest of the file is unchanged.
~/.qoderwork/skills/felo-youtube-subtitling/A small bundle. The valuable part is the caption fetch, and the SKILL.md explains exactly what it returns.~/.qoderwork/skills/The root it matches from. QoderWork picks a skill by its description, so a short list makes the matching predictable rather than clever.Which skill wrote to my notes file, and from which video?The transcript on its own is an intermediate file. What matters is that the notes end up in your document with the quotations anchored, so anyone reading them later can check the source in a minute.
We spent the first year on the wrong problem, and the way we found out was that nobody used the reporting screen.
Adoption went from eleven percent to sixty after we moved the same feature into the export flow.
So the feature was fine, the placement was wrong.
Placement was the whole thing. We had shipped it twice before without noticing.
Nobody asked for a new dashboard. Three people asked to take the numbers with them.
Placement was the whole thing. We had shipped it twice before without noticing.
The difference between a useful note and a forgotten one is whether the quotation can be checked later.
Somebody shares a talk, a webinar, or a recording, and the useful part is four sentences long.
Ask for the text, not a summary. In a local workflow this is cheap: the transcript is just another file on your machine, and you can search it later.
Have the notes added to your document with a timestamp beside each quotation. In six months the timestamp is the only thing that will let you verify the claim.
A quote without the sentence before it can be made to say the opposite. Ask for both.
Either keep the full text where you can search it or drop it once the notes are written. Leaving it in the documents folder is how folders become unusable.
Leaves behindA few quoted, timestamped points added to the notes file you already keep, with the surrounding context preserved.
Both produce something in a local file rather than a reply you have to remember.
Get the transcript of this talk and add the three most useful points to ~/Documents/reading-notes.md with timestamps.What comes backYour notes file updated with anchored quotations, in the format you already use.
Pull every figure from this webinar into a table in notes.md, with the minute it was said.What comes backA checkable table in the file, which is more useful than a summary for anything with numbers in it.
Captions are convenient and imperfect, and the imperfections are predictable.
Auto-captions are good enough to search and not good enough to quote blind. Verify any figure that will be repeated in a document somebody else signs off on.
A recording without subtitles gives the skill nothing to fetch. Recordings behind a login are out of reach entirely.
Notes for internal use are one thing; repeating a speaker's words in a published document is a different decision, and not one the skill makes.
Everything here is on your own machine, so verification takes a minute rather than a meeting.
Open the video at one of them. A timestamp that points at silence means the alignment slipped and the quotations cannot be trusted.
Check the document you named. A note-taking task that edits other files is working outside the request.
Read one quotation with the sentence before it. If the meaning flipped, the extraction is too aggressive.
Five skills written for QoderWork. They follow the same pattern on this platform, so you can combine them in one session without re-explaining your setup.
Start with a skill, then compose the exact models, live data, knowledge, and output tools your QoderWork workflow needs — the same key every Felo skill already uses.
Use cost-efficient LLM access when your agent needs custom reasoning or generation.
Give agents current web and social signals instead of relying on stale context.
Ground work in reusable knowledge, then turn it into shareable deliverables.
Extend the workflow from research to visual and office-ready results.