How to use the Media Ingest agent skill (media-ingest)

What an agent skill is and what media-ingest makes the agent do

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a particular kind of task. The agent reads these instructions when a request matches the skill.

The media-ingest skill covers video, audio, PDF, book, screenshot and GitHub repo content. For each format it fetches the material: a transcript for video or audio (from speech-to-text, a transcription service or captions), extracted text for PDFs (with OCR if needed), OCR via a vision model for screenshots, or cloned repo files for GitHub repos. It then creates a brain page with a summary rather than a raw transcript dump.

• Saves the original file for provenance using gbrain files upload-raw
• Files the page by primary subject, not by media format
• Finds people and companies mentioned and adds back-links from their pages to the new item
• Adds timeline entries on those entity pages
• Runs gbrain sync at the end to update the index

Install the media-ingest skill

Download the ZIP for the skill from this page. The skill is a single file, SKILL.md, inside the plugin-variants/gbrain-daily/skills/media-ingest folder of the garrytan/gbrain repository.

Option A, personal install: unzip the download into ~/.claude/skills/ so the skill folder sits there as a personal skill.

Option B, project install: unzip it into the .claude/skills/ folder of a specific project.

Option C, Claude apps: open Settings, go to Capabilities, then Skills, and upload the ZIP there.

Check that the skill folder contains SKILL.md after unzipping.

Prerequisites: what you need before it works

The skill depends on a working gbrain installation that provides its brain, search, page and file upload tools. Without these, the skill cannot write pages or store raw files.

It also needs a speech-to-text, caption, or OCR/vision service. The skill names these generically; which provider you use and which API keys you need depend on your own setup. The source does not name specific providers.

The skill refers to two other pieces that are not included in the files reviewed: the companion enrich skill, used to create or update entity pages, and a filing rules file (skills/_brain-filing-rules.md). Check that both are available before relying on the skill.

How to trigger the skill with natural requests

You do not need a special command. Ask in plain language, and the agent will match your request to the skill. The skill's listed trigger phrases include the examples below.

Give the agent a source it can reach, such as a URL or a file path, or a file you have uploaded. You can add an optional title or target slug to override the auto-detected values.

• "Watch this video and add it to my brain: [YouTube link]"
• "Process this YouTube link" followed by the URL
• "Ingest this PDF" with the file attached
• "Save this podcast" with the audio link
• "Process this book" or "summarize this book" for a PDF book
• "What's in this screenshot?" with the image attached
• "Check out this repo" with a GitHub link
• "Ingest it into my brain" after sharing content

Worked example: ingesting a video and a book

Example one: you paste a YouTube link and say "process this YouTube link". The agent fetches the transcript from captions or a transcription service, writes a page with a summary and timestamped highlights, saves the raw file with gbrain files upload-raw, and checks whether any speakers have existing brain pages. If a speaker has a page, it adds a back-link and a timeline entry to that page.

Example two: you attach a PDF book and say "process this book". The skill identifies chapters or sections and writes chapter summaries. For books over 500 pages it summarises by chapter instead of inlining the full text, and it links to the raw upload.

Safety notes: scripts, network and sensitive content

The skill does not run scripts, but it does use the network. It fetches content from URLs, GitHub repos and transcription or vision services, and it writes and updates multiple pages and links in your knowledge base.

Content is fetched and written into a stored knowledge base and may pass through transcription or vision services. The source does not say which services are used, so do not send confidential or sensitive documents unless you have checked where they go.

If you only want a quick one-off summary and do not want your knowledge base changed, do not use this skill.

• Do not use it for sensitive or confidential documents unless you have checked the services involved
• Expect multiple page writes and links per item
• Check file_upload limits: uploads refused for size or path reasons are not stored, and the page keeps a pointer instead

Tips for good results

Scanned books with poor OCR often come out garbled. The skill checks OCR quality and flags items that are less than 80% readable instead of ingesting them, so read those flags rather than trusting the page.

Video and audio with several speakers but no speaker diarization get a prominent limitation note, because speech cannot be reliably attributed to each person. YouTube auto-captions can misspell names, so the skill cross-checks names against existing pages first.

Do not re-run ingest on the same source without thinking. The skill checks for an existing source URL or file hash and asks whether to update or skip, but duplicates are a known pitfall if you bypass that check.

Audio longer than about two hours may time out, so the skill can split it into chunks first.