Cowork + Skills: Automating Your Work

Terminal-based. Deep code editing, file operations, Git, shell commands, debugging. Built for developers who live in the terminal.

Desktop app. Skills system, plugins, Desktop Commander, scheduled tasks, GUI-friendly. Built for writing, research, analysis, and cross-tool automation.

mkdir -p ~/.claude/skills/meeting-notes-extractor

Create ~/.claude/skills/meeting-notes-extractor/SKILL.md:

# Meeting Notes Extractor Skill

## When to Use
Trigger when user says: "extract meeting notes", "clean up these notes",
"process my meeting notes", "format this meeting summary", or pastes
raw meeting notes and asks for a summary.

## What You Need
No MCPs required. Works on text provided directly by the user,
or a file path they specify.

## Instructions

### Step 1: Parse the Raw Input
Read the raw notes provided. Identify:
- Meeting title / project (from context or first line)
- Date (from notes or ask user)
- Attendees (look for names, @mentions, or "Present:" sections)
- The overall meeting goal/agenda

### Step 2: Extract Key Elements
Scan the full notes for:
- DECISIONS: Any statement where a choice was made or direction agreed
  (look for: "we decided", "agreed to", "going with", "confirmed", "approved")
- ACTION ITEMS: Any task assigned to a person with implied deadline  
  (look for: names + will/should/needs to/by [date], "AI:" prefix, "TODO:")
- OPEN QUESTIONS: Anything unresolved, tabled, or needing follow-up
  (look for: "TBD", "need to check", "not sure", "revisit", "?")
- KEY INSIGHTS: Important points, discoveries, or context worth remembering

### Step 3: Format Output

Use exactly this template:

---
## Meeting: {title}
**Date:** {date} | **Attendees:** {list}

### 🎯 Decisions Made
{bullet list — each decision as one clear sentence}

### ✅ Action Items
| Owner | Action | Due |
|-------|--------|-----|
{row per action item — use "TBD" for unspecified owners/dates}

### ❓ Open Questions
{bullet list — each question with who owns finding the answer}

### 💡 Key Insights
{bullet list — 3-5 notable points worth remembering}

### 📋 Next Steps
{1-3 sentence summary of immediate next steps}
---

### Step 4: Offer to Save or Send
After outputting the summary, ask:
"Would you like me to: (a) save this to a file, (b) create a calendar 
follow-up, or (c) draft a follow-up email to attendees?"

## Quality Check
Before outputting, verify:
- Every action item has an owner (even if "TBD")
- No decision was missed (re-read the raw notes once more)
- The output is clean and could be sent directly to attendees

## Example Input
"call w/ sarah and mike re: API redesign. we decided to deprecate v1 
endpoints by march. sarah will write migration guide by jan 15. 
mike handles backward compat shim. still tbd: pricing for enterprise 
tier. action: follow up with legal re: data retention policy (me, asap)"

## Example Output
Meeting: API Redesign Discussion
Date: [today] | Attendees: Sarah, Mike

🎯 Decisions Made
- Deprecate v1 API endpoints by March (date TBD)
- Implement backward compatibility shim during transition period

✅ Action Items
| Owner | Action | Due |
|-------|--------|-----|
| Sarah | Write migration guide | Jan 15 |
| Mike | Build backward compat shim | TBD |
| [You] | Follow up with legal on data retention policy | ASAP |

❓ Open Questions
- Enterprise tier pricing model — who owns this decision?

💡 Key Insights
- v1 deprecation requires a migration path for existing customers
- Legal review needed before finalizing data retention approach

In Cowork, paste this raw text and see if the skill triggers:

Process my meeting notes:
"standup with team. john will fix the login bug by friday. 
we agreed to push the feature launch to next sprint. 
sarah asked about the db migration - still need to check with ops. 
everyone should update their tickets by eod."

Claude should recognize the trigger, follow the SKILL.md instructions, and output a clean structured summary.

Look at what the skill produced. Common first-pass issues to fix in your SKILL.md:

• Action items missing an owner → add more "look for" patterns to Step 2
• Decisions not captured → add more trigger phrases to the DECISIONS section
• Output format wrong → adjust the template in Step 3
• Skill not triggering → add more phrases to "When to Use"

Two or three iterations of this produces a skill that works reliably every time.

Claude asks you 5-7 questions:

• "What triggers this skill? What would you say to start it?"
• "What does the skill need to do, step by step?"
• "What tools or MCPs does it need access to?"
• "What should the output look like? Show me an example."
• "Are there things it should never do?"

Based on your answers, Claude generates a complete SKILL.md, saves it to the right location, and tests it immediately with a sample input you provide.

You: My Downloads folder is a mess. Organize it by file type into 
subfolders (PDFs, Images, Documents, Archives, Other).

Claude uses Desktop Commander to:
1. list_directory("/Users/yourname/Downloads")
2. Categorize each file by extension
3. create_directory for each category that doesn't exist
4. move_file each file to the right subfolder

Result: 847 files organized in 23 seconds.

You: Rename all the screenshots in ~/Desktop that start with 
"Screenshot 2024" to use the format "2024-MM-DD-HH-MM.png"

Claude reads the directory, parses the existing filenames, 
maps them to the new format, and renames each one.

You: In ~/Documents/reports/, I have 12 monthly CSV files.
Read each one and give me a summary: total rows, date range,
and any rows where the "status" column is "error".

Claude reads all 12 files, analyzes each, and returns a 
consolidated summary table.