Guide · Agent skill
How to use the Meeting Transcript Ingestion agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The meeting-ingestion skill is an agent skill for gbrain, a page-based personal knowledge base. It takes a meeting transcript from a recorder or pasted text and produces a structured meeting page, plus linked pages for attendees and companies. This guide is for people who record meetings and want a searchable, linked record of who was there and what was decided. After reading it you will know how to install the skill, how to ask for it in plain language, what it needs, and where its limits are.
Meeting Transcript Ingestiongarrytan/gbrain · MIT Download the skill Quick start
- 1Download the skill ZIP. Get the ZIP from the skill page on this site. It contains the SKILL.md file for meeting-ingestion.
- 2Install it in the right folder. Unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
- 3Confirm your gbrain setup. Check that your gbrain knowledge base provides the listed tools and that the companion files are available.
- 4Ask in plain language. Paste the transcript or point the agent to the export and ask it to process the meeting or ingest the recording.
- 5Review the result. Check the meeting page, the attendee and company pages, and the source transcript page. The meeting is only reported as ingested once the verification checklist passes.
What an agent skill is and what this one 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. Once installed, the agent can load the skill when a request matches its triggers.
When this skill is installed, the agent processes a meeting transcript from any recorder, including an AI notetaker export, a webhook, pasted text or an audio transcript. It writes a structured meeting page into your brain (the gbrain knowledge base). It also splits recordings that contain several meetings, identifies speakers by evidence, creates or updates pages for attendees and companies, adds timeline entries and two-way links, and checks any recorder summary against the transcript before trusting it.
- Meeting page with attendees, summary, key decisions, action items and notable quotes
- Pages under meetings/, people/ and companies/
- Timeline entries and back-links on every mentioned entity
- A verification checklist that must pass before the meeting is reported as ingested
Install the meeting-ingestion skill
Download the ZIP for the skill from the skill page on this site. The skill is distributed as a folder containing SKILL.md, so unzip the download so that the skill folder ends up in one of the locations below.
For a personal install, unzip it into ~/.claude/skills/. To limit it to one project, unzip it into the .claude/skills/ folder inside that project. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Single project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check the requirements before you run it
The skill depends on a gbrain knowledge base. It calls the tools search, query, get_page, put_page, add_link, add_timeline_entry, get_timeline and chronicle_day. Without a gbrain or a compatible page-based knowledge base, the skill has nothing to write to.
The skill also refers to companion files, _brain-filing-rules.md, conventions/quality.md and conventions/brain-first.md. These are not included in the SKILL.md file alone, so make sure they are available in your setup. You also need access to the recorder export or the pasted transcript.
- A gbrain knowledge base with the tools listed above
- The companion files _brain-filing-rules.md, conventions/quality.md and conventions/brain-first.md
- The meeting recorder export or the pasted transcript text
Trigger the skill with natural requests
You do not need a special command. Ask in plain language, and the agent will load the skill when your request matches one of its triggers. The triggers include phrases such as meeting transcript, process this meeting, meeting notes, meeting recorder, ingest this recording, capture meetings, audit this meeting, check the sequence and did I get the order right.
Give the agent the transcript, either as an export file or pasted text, along with the request. A recording that contains several meetings is split first, so you do not need to separate them yourself.
- "Process this meeting: [paste transcript]"
- "Ingest this recording and file the attendees and decisions."
- "Audit this meeting and check the sequence of events."
- "Did I get the order right in this call?"
What happens during an ingestion run
The skill first reduces any input to a normalized transcript record that holds the source, attendees, segments and the complete raw transcript. It then checks whether one recording holds several meetings and splits it if so. Speakers are resolved by evidence, never by guess, and any surprising claim goes through a consistency check against the transcript and the brain.
Every attendee and company gets a page or an update, and every mention gets a back-link to the meeting page. A meeting is not reported as ingested until the verification checklist passes, including the sequence check. If the sequence check finds a contradiction, ingestion is blocked until it is fixed or you explicitly waive it.
- Normalize the input and stop if the transcript is empty or only a summary
- Split multi-meeting recordings into separate pages
- Resolve speakers by evidence and flag any that stay unresolved
- Check surprising claims against the transcript and brain
- Enrich every entity and run the verification checklist
Safety notes: scripts, network and retained transcripts
The skill is listed as running no scripts and using no network. However, the recorder or service that produced the transcript may send data over a network, which is outside this skill. Check that before you send a confidential meeting through a recorder.
Raw transcripts are kept as source pages at sources/meetings/ with type source and facts_backstop set to false. Before that file is written, the skill scans for secret-shaped strings and personal data and replaces matches with labeled placeholders. Review the pages it creates, because anything you do not review is kept in your knowledge base.
- No scripts and no network use are declared for this skill
- Raw transcripts are stored as source pages, so review them
- Secrets and PII are redacted before the sidecar is written
- Do not use it for sensitive meetings if the transcript goes to a service you do not trust
Tips for better results
Give the most complete transcript you have. The skill never truncates the raw transcript, but a partial export will produce a partial record. Mark a recording as in progress only when it is finished, because in-progress recordings are stopped at the start.
If you distrust a recorder's AI summary, say so. The skill treats that summary as a claim and checks it against the transcript, so you can rely on quotes that the verification step has confirmed.
- Use the full export, not a summary-only payload
- Wait until the recording is finished before ingesting
- Review the meeting page and the source transcript page after each run
What people use it for
- Filing a weekly client call so that each attendee and company page gains a dated timeline entry
- Recording decisions and action items from a project meeting into a searchable meetings/ page
- Auditing a recorder's AI summary by checking its claims against the raw transcript
- Splitting a recording that ran across several back-to-back sessions into separate meeting pages
Frequently asked questions
Is the meeting-ingestion skill free to use?
The skill is published under the MIT licence. The licence terms and the cost of your gbrain knowledge base and any recorder you use are separate questions, and this page does not state prices.
Do I need gbrain to use this skill?
Yes. The skill writes to meetings/, people/ and companies/ pages and calls gbrain tools such as search, get_page and put_page. Without a gbrain or a compatible page-based knowledge base, it is not suitable.
How do I install meeting-ingestion?
Download the ZIP from the skill page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills.
Does the skill send my transcript to the internet?
The skill is listed as using no network and running no scripts. The recorder or service that produced the transcript may still send data over a network, so check that before ingesting confidential meetings.
What if the skill says a meeting is not finished?
The skill will not report a meeting as ingested until its verification checklist passes, including the sequence check. A sequence contradiction blocks ingestion until you fix it or explicitly waive it.
How does it compare to a simple meeting summary tool?
A simple summary tool gives you a one-off summary. This skill runs a multi-step pipeline with checks and writes linked pages into a knowledge base, which is heavier than needed for a quick one-off summary.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you record meetings, already use or will set up a gbrain knowledge base, and want attendees, decisions and quotes checked against the raw transcript. Pick something lighter if you only need quick one-off summaries, or if you do not want a knowledge base at all, or if the meeting is too sensitive to store as reviewed source pages.
Download the skill