Incredible!
Let’s all repeat it… Incredible!
After a few weeks from the announcement, I’m finally trying Copilot Cowork.
The idea is to create a skill (yes, the same kind highlighted in Anthropic tools) that supports users in a meeting by reading all the information, generating a recap, and saving it autonomously.
All without lifting a finger!
Here’s what I did:
I created my “Meeting Action Plan” folder in OneDrive, inside the Cowork -> Skills folder, containing the SKILL.md file.
What is the SKILL.md file?
It’s a de facto convention widely used by people building Agents, MCP tools, or AI-driven repositories.
In short
SKILL.md is a Markdown file that describes an agent “skill”, i.e., a well-defined capability, with rules, inputs, outputs, and clear limits, designed to be:
- read well by an LLM
- versionable
- composable with other skills
- loaded as “structured” context
A skill is:
- bigger than a prompt
- smaller than a full agent
- self-documented
- reusable
You can think of it as a “cognitive function” described in natural language, not in code.
Next step
Back to us.
The .md file must follow conventions and a clear structure, and I deliberately do exactly the opposite.
Once the file is created, I open it and write the following text:
## Post-Call Meeting Intake (Recap Automation)
_A skill to collect, in a structured way, all the information from a call that has just ended and generate a complete recap, ready to be shared._
### Overview
This skill guides the user, right after a call/meeting, through collecting all the essential information needed to automate the creation of minutes and a recap. The goal is to drastically reduce manual work (notes, follow-ups, summary emails) and standardize the quality of the output.
This skill is designed for:
– Project managers, consultants, account managers, team leads, and anyone who needs to send recurring recaps
– Internal and external meetings (clients, vendors, stakeholders)
– Operational calls, steering committees, workshops, technical alignments, UAT sessions, etc.
The expected outcome is a structured dataset + a “ready to send” recap that includes: context, agenda, discussion points, decisions, actions with owners and due dates, risks/blocks, open questions, and next steps.
### Key Points
– Start the skill immediately after the call and collect the basic metadata (title, date/time, duration, participants, client/project, channel).
– Ask for a structured summary of the content: call objective, agenda, topics covered, results achieved, and critical points.
– Extract and normalize “Decisions”, “Actions”, and “Open Questions” into clear lists (with owner, due date, priority, dependencies, and status).
– Capture risks, blockers, and assumptions that emerged, including impacts and proposed mitigations.
– Produce a final recap in a shareable format (text for email/Teams) + an immediate, actionable “Next steps” section.
– If information is missing, ask targeted questions (max 1 at a time) to complete the essential fields without slowing the user down.
– Apply a quality standard: avoid duplicates, highlight inconsistencies, flag missing information, and propose a “clean” version of the recap.
– Recommended Output:
1) Executive summary (5–8 rows)
2) Details
3) Decisions
4) Action items (owner + due date)
5) Risks/Blocks
6) Open questions
7) Next call / follow-up
_Last updated: 2026-04-25_
I save it and go to Copilot Cowork, and this is where things get really interesting—maybe precisely because of the Anthropic models used “under the hood”.
- First question:
“I just created a skill.md file, can you access it?”
Answer:
“Yes, I can see it now. I read your skill ‘Post-Call Meeting Intake (Recap Automation)’ from OneDrive — 2210 bytes, updated today.”
In short, it’s a skill that, right after a call, collects metadata (title, date, participants, client/project), a structured summary (agenda, topics, outcomes), decisions, action items with owner/due date, risks, open questions, and produces a recap ready to send via email or Teams.
Do you want me to register it as an active personal skill in your profile, or are we just reviewing it? I can also suggest improvements to the text before saving it.”
- Second question:
“Is it formatted correctly, does everything seem fine?”
Answer (⚠️ Pay attention here):
“No, the most important part is missing: the YAML frontmatter. Without it, the skill won’t be loaded by the system.
What’s missing at the beginning of the file:
yaml
—
name: meeting-action-plan
description: |
Collects structured information right after a call and generates a recap
ready to share. Use when the user asks to “do the call recap”,
“generate the meeting minutes”, “collect action items after the meeting”,
“prepare the call follow-up”.
—
What’s good:
Clear, well-organized content (Overview, Key Points, output structure)
Well-defined final output (7 sections)
Explicit use cases
To fix in addition to the frontmatter:
The H2 title (“Post-Call Meeting Intake”) doesn’t match the folder name (“Meeting Action Plan”) — it’s better to align them.
Overlap with meeting-intel: the system skill already covers recaps, action items, and decisions from transcripts. It’s worth clarifying what yours does differently (e.g., manual collection via questions vs. automatic extraction from transcripts), otherwise both may end up triggering.
A “When to Use” section with explicit trigger phrases helps routing — yours describes scenarios but without examples of user prompts.
Do you want me to fix it and save the corrected version to OneDrive?
- Third question:
“Fix it and translate it all into English.”
Let’s just say that, after a few commands/prompts, the .md file gets configured correctly.
Now that file analyzes meetings, produces a recap, saves it as a Word file, and asks the user where to save it.
Inside the Cowork chat, within a few moments, the .md file is corrected for me and uploaded to the desired folder.
At this point I select a meeting where I recorded and enabled Copilot transcription.
Here’s what I ask—still inside Cowork:
“Can you give me a list of the meetings from the last 2 weeks that I can run the Meeting Action Plan on?”
Copilot shows me a set of meetings I can act on. I select one and ask it to proceed with the recap.
On screen, it shows me the recap… and up to this point, Copilot 365 could do that too.
Oh, if you’re interested, I published a prompt that’s suitable for retrieving all information after a meeting.
Prompt after Meeting – Copilot M365 – BEYOND THE PLATFORMS
Here’s the magic: by following the instructions in the .md file, Copilot generates the Word file for me and saves it—or, for example, it could send it to our colleagues.
From now on, every time I enter Cowork, I’ll just need to type this prompt to activate my skill:
“Generate the Meeting Action Plan for the meeting [meeting name]”
Noteworthy features.
- Cowork lets us submit multiple requests that are queued one after another and handled First In – First Out.
- I can pause the process or suspend it and then reactivate it with commands (Pause (soft), Pause (hard), Resume).
- I can cancel the current process with the “Cancel” command.
These are some of the things you can do.
The feeling, using Copilot Cowork this way, is clear:
we’re not writing better prompts—we’re designing behaviors.
SKILL.md isn’t meant to “tell Copilot what to answer”, but to teach it what to do, when to do it, and with what quality standard.
It’s the shift from a reactive assistant to an operational colleague, capable of owning an outcome from start to finish.
Today it’s a meeting recap.
Tomorrow it could be any repetitive, delegable process that consumes a lot of attention.
Copilot Cowork isn’t a new way to use AI.
It’s a new way to design work.
Boom, done 💣!
Follow me:
BEYOND THE PLATFORMS 
Leave a comment