Meeting Summary Skill

Transforms raw meeting transcripts into clean, structured, actionable summaries.

Input

The transcript may arrive as:

Pasted text directly in the chat
An uploaded file (check /mnt/user-data/uploads/ — read the file-reading skill if needed)
A partial transcript with speaker labels, timestamps, or both If the transcript is missing, ask the user to paste it or upload the file before proceeding.

Output Structure

Produce the summary in this order. Use markdown headers. Keep each section tight — prioritize signal over completeness.

1. 📋 Meeting Overview

Field	Value
Date	(extract from transcript or write "Not specified")
Attendees	(names or roles mentioned)
Duration	(if inferable)
Purpose	(1-sentence statement of why the meeting happened)

2. 🗝️ Key Decisions

Bullet list. Only include actual decisions made — not discussions, not options considered. If no decisions were made, write "No formal decisions recorded."

3. ✅ Action Items

Table format:

#	Action	Owner	Due Date	Priority	Status
1	...	Name / Role	Date or "TBD"	🔴 High	Open

Priority definitions — infer from context in the transcript:

🔴 High — blocking, urgent, time-sensitive, or explicitly called out as critical / must-do
🟡 Medium — normal delivery, expected as part of regular work
🟢 Low — optional, nice-to-have, or flagged as "if time allows" / "eventually" When priority is not explicitly stated, infer it from urgency signals (e.g., "ASAP", "before the launch", "blocking the team" → High; "whenever you can", "no rush" → Low; everything else → Medium).

Status definitions:

Open — not yet started or in progress (default for all new action items from the transcript)
In Progress — explicitly mentioned as already underway
Done — explicitly confirmed as completed during the meeting If no owner or date was mentioned, write "TBD". If no action items exist, write "No action items identified."

4. 📝 Summary of Discussion

3–6 bullet points covering the main topics discussed. Each bullet = one topic or theme. Write in plain language — no jargon unless it appeared in the transcript.

5. 🔗 Follow-up & Open Questions

Bullet list of unresolved questions, items deferred to future meetings, or things explicitly flagged as "to be confirmed / decided later." If none, omit this section.

Processing Guidelines

Speaker attribution: If the transcript has speaker labels (e.g., "John:", "Speaker 1:"), use names when attributing decisions and action items. If no labels, attribute to roles if inferable (e.g., "the PM", "the client").
Timestamps: Ignore unless they help resolve ambiguity.
Filler/small talk: Strip it. Focus on substance.
Repetition: Deduplicate. If the same point was made 3 times, summarize it once.
Confidential signals: If the transcript includes markers like "[REDACTED]" or "[OFF THE RECORD]", respect them — do not infer or fill in gaps.
Long transcripts (>5,000 words): Process in logical segments (opening, main discussion, closing) and synthesize across them. Do not truncate.
Multiple agenda items: Add a sub-header under "Summary of Discussion" for each agenda item.

Tone & Style

Neutral, professional tone
Past tense ("The team decided...", "Sarah agreed to...")
Concrete and specific — avoid vague summaries like "the team discussed various topics"
If something is unclear in the transcript, note it with (unclear in transcript) rather than guessing

Output Format Options

Default: Markdown in chat.

If the user asks for a Word document or downloadable file, read the docx skill and produce a .docx file in /mnt/user-data/outputs/.

If the user asks for a PDF, read the pdf skill instead.

Example Trigger Phrases

"Here's the transcript from today's standup, can you summarize it?"
"Summarize this call recording"
"Extract the action items from this meeting"
"Write up the notes from this conversation"
"What did we decide in this meeting?"
(User pastes a wall of text that looks like a transcript)