meeting-video-grounding
Skill444 starsChanged 7 months agoMITView on GitHub
Convert a meeting video into an audio-first transcript bundle, then use meeting-grounding to produce structured meeting grounding outputs.
In this repository15 other files it gives its agents.
- archive-groundingSkill
- audio_structuringSkill
- document-groundingSkill
- grounded-research-litSkill
- grounded-reviewSkill
- grounded-summarySkill
- input-routerSkill
- meeting-audio-groundingSkill
- meeting-groundingSkill
- one-reportSkill
- pptx-groundingSkill
- remote-inputSkill
- report-exportSkill
- skill-evolveSkill
- table-groundingSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
22 rules. About 0 are common, 17 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈17
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
Use this skill when:
- Distinctive: line 28
Do not use this skill when:
- Distinctive: line 70
If the meeting contains multiple independent topics that should be researched separately downstream, the bundle may also contain:
- Distinctive: line 96
always writing the meeting-level:
- Very specific: line 107
Run the existing scripts/run.sh entrypoint for this skill.
- Very specific: line 113
Read transcript/meeting_transcript.txt as the primary grounding evidence.
- Very specific: line 115
Always save the meeting-level structured note to grounded.md inside the same bundle directory.
- Distinctive: line 116
If the transcript clearly contains multiple independent topics, also save:
- Distinctive: line 119
Do not stop after confirming that the transcript bundle exists.
- Distinctive: line 129
do not invent slide content, visual details, or screen evidence that were not captured in the transcript
- Very specific: line 135
If topic children are created, each child grounded note must also follow the same meeting-grounding schema.
- Distinctive: line 137
Do not invent a new schema here.
- Distinctive: line 141
Do not implement a new ASR pipeline.
- Distinctive: line 142
Do not implement a new meeting summarizer.
- Distinctive: line 143
Do not directly summarize the raw video without first running the existing workflow.
- Distinctive: line 146
Treat the transcript as the primary evidence.
- Distinctive: line 147
Keep the workflow simple and stable.
- Distinctive: line 151
If the input video file does not exist, fail clearly.
- Distinctive: line 152
If the video has no audio stream, fail clearly.
- Distinctive: line 153
If audio extraction fails, fail clearly.
- Very specific: line 154
If meeting_transcript.txt is not produced, fail clearly.
- Distinctive: line 155
Do not pretend the task succeeded if only part of the workflow completed.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Meeting Video Grounding
Convert a meeting video into structured meeting grounding outputs by:
- extracting audio from the video
- reusing the existing
audio_structuringskill to producemeeting_transcript.txt - reusing the existing
meeting-groundingskill to turn that transcript into meeting grounding outputs
This skill is for meeting videos where the primary information comes from speech. It is intentionally audio-first. It does not attempt full visual understanding of the video.
When to Use
Use this skill when:
- the input is a meeting video or discussion video
- the main information is expected to come from spoken content
- you want to reuse the existing audio transcription and meeting grounding workflow
Do not use this skill when:
- the task requires visual analysis of slides, demos, whiteboards, or screen content as first-class evidence
- the video has little or no speech
- the task is to write a polished final report directly from the raw video
Input
A single meeting video file.
Typical examples:
.mp4.mkv.mov.webm
Optional input:
transcription_language: language code such asenorzh
Output Bundle
For each input video, create one bundle directory:
data/grounded_notes/<ground_id>/Inside that bundle, the expected outputs are always:
<bundle_dir>/
├─ extracted.md
├─ extracted_meta.json
├─ grounded.md
├─ audio/
│ └─ meeting_audio.wav
└─ transcript/
└─ meeting_transcript.txtIf the meeting contains multiple independent topics that should be researched separately downstream, the bundle may also contain:
<bundle_dir>/
├─ topic_manifest.json
└─ child_outputs/
├─ topic_01/
│ └─ grounded.md
├─ topic_02/
│ └─ grounded.md
└─ ...Important separation of responsibilities
scripts/run.shis responsible for:- extracting audio from the input video
- calling the existing
audio_structuringskill - creating the bundle files:
audio/meeting_audio.wavtranscript/meeting_transcript.txtextracted.mdextracted_meta.json
- The agent is responsible for:
- reading the transcript bundle
- applying the existing
meeting-groundingskill - always writing the meeting-level:
grounded.md
- and, when appropriate, also writing:
topic_manifest.jsonchild_outputs/topic_xx/grounded.md
grounded.md must be a real grounding note.
It must not remain a placeholder scaffold.
Required Agent Workflow
- Run the existing
scripts/run.shentrypoint for this skill. - Confirm that the bundle exists and that these files are present:
extracted.mdextracted_meta.jsonaudio/meeting_audio.wavtranscript/meeting_transcript.txt
- Read
transcript/meeting_transcript.txtas the primary grounding evidence. - Reuse the existing
meeting-groundingskill on that transcript. - Always save the meeting-level structured note to
grounded.mdinside the same bundle directory. - If the transcript clearly contains multiple independent topics, also save:
topic_manifest.jsonchild_outputs/topic_xx/grounded.md
- Do not stop after confirming that the transcript bundle exists.
Important scope rule
This skill currently treats meeting videos as audio-first inputs.
That means:
- the extracted transcript is the primary evidence for grounding
- absence of visual analysis is not, by itself, a failure
- do not invent slide content, visual details, or screen evidence that were not captured in the transcript
Output Format
The final grounded.md must follow the existing meeting-grounding schema exactly.
If topic children are created, each child grounded note must also follow the same meeting-grounding schema.
Do not invent a new schema here.
Instructions
- Do not implement a new ASR pipeline.
- Do not implement a new meeting summarizer.
- Do not directly summarize the raw video without first running the existing workflow.
- Reuse the existing
audio_structuringskill for transcription. - Reuse the existing
meeting-groundingskill for transcript grounding. - Treat the transcript as the primary evidence.
- Keep the workflow simple and stable.
Failure Handling
- If the input video file does not exist, fail clearly.
- If the video has no audio stream, fail clearly.
- If audio extraction fails, fail clearly.
- If
meeting_transcript.txtis not produced, fail clearly. - Do not pretend the task succeeded if only part of the workflow completed.
Example Invocation
/meeting-video-grounding
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

