How Skills Work

Skills are reusable instruction sets that tell Claude how to handle specific tasks. Instead of explaining what you want every time, you write it once and invoke it by name.

---

Skill Matching

When you type a message, Claude does not just respond — it first checks whether any skill matches your request.

The matching process:

1. You say something: "process my meeting from this morning"
2. Claude scans the YAML headers of all installed skills
3. It finds a match: the process-meeting skill has a trigger for "process meeting"
4. It loads the full skill body (the detailed instructions)
5. It follows those instructions to handle your request

This happens automatically. You do not need to think about which skill to use — you just describe what you want and Claude figures out which skill applies.

What if no skill matches? Claude responds normally, using its general knowledge plus whatever is in your CLAUDE.md config. Skills are enhancements, not requirements. Your KB works without any skills installed — they just make common tasks faster and more consistent.

---

Slash Commands vs Natural Language

You can invoke skills two ways:

Slash Command (Explicit)

/process-meeting

This directly invokes the skill. No ambiguity, no matching needed.

Natural Language (Implicit)

Process my meeting from this morning

Claude matches this to the process-meeting skill based on the trigger words.

Both work. Slash commands are faster when you know exactly which skill you want. Natural language is easier when you are just talking to Claude and the right skill fires in the background.

You can even mix them:

/process-meeting — it was a 1:1 with Elaine about the gateway migration

The slash command triggers the skill, and the rest of your message provides context for it.

---

What Makes a Skill

A skill is a Markdown file with two parts:

1. Frontmatter (YAML Header)

---
name: process-meeting
description: Transform meeting transcripts into structured notes
  with extracted decisions, action items, and stakeholder positions.
triggers:
  - "process meeting"
  - "process my meeting"
  - "meeting notes from"
  - "process this transcript"
version: 1.2
---

The frontmatter is loaded at the start of every conversation (headers only — not the full file). This is how Claude knows the skill exists and what triggers it.

2. Body (Instructions)

The body contains the detailed instructions Claude follows when the skill fires. This is plain Markdown with step-by-step directions:

## Instructions

When processing a meeting transcript:

1. Detect the meeting type (standup, 1:1, project sync, etc.)
2. Extract all attendees and validate against People/ profiles
3. Write a structured summary with these sections:
   - Quick Summary (2-3 sentences)
   - Key Discussion Points
   - Decisions Made
   - Action Items (with owners and deadlines)
4. Extract signals to relevant People profiles
5. Update Work Board with new action items
6. Save to Meetings/ with date-based filename

## Extraction Rules

For each attendee, extract:
- Positions expressed (what they advocated for)
- Concerns raised (what worried them)
- Commitments made (what they agreed to do)
...

The key principle: Headers are always loaded (lightweight). Bodies are loaded on demand (only when the skill fires). This means you can have dozens of skills without bloating every conversation.

---

Skill Chains

Some tasks naturally involve multiple skills. Rather than making you invoke each one, skills can trigger other skills in sequence.

Example: Processing a meeting triggers a chain

You: "process my meeting"
  │
  ├── process-meeting (primary skill)
  │     Writes the structured meeting note
  │
  ├── extract-signals (chained)
  │     Pulls out positions, concerns, commitments
  │     Updates People profiles
  │
  ├── track-deliverables (chained)
  │     Adds action items to Work Board
  │     Sets deadlines and owners
  │
  └── scan-for-content (chained)
        Checks if any discussed topics need
        Decision records or Project updates

You said one thing. Four skills fired. Your meeting note is written, People profiles are updated, your Work Board has new items, and a decision record was created — all from a single request.

Not all skills chain. Simple skills like "file this document" or "add a note" run alone. Chains are for complex workflows where multiple extractions happen from one input.

---

Starter Kit

When you set up your KB, you start with a core set of skills. These handle the most common tasks:

Skill	What it does
process-meeting	Transforms transcripts into structured notes with extractions
file	Files a document into the right KB folder with metadata
note	Creates a quick note in Inbox/ for later filing
start-day	Morning routine: reviews calendar, surfaces priorities, shows Work Board
end-day	Evening routine: logs what happened, moves items on Work Board
week-review	Weekly review: summarizes the week, identifies patterns, preps next week

These six skills cover probably 80% of daily KB usage. You can use your KB for months with just these.

What each one unlocks:

• process-meeting is the workhorse. Most of your KB content comes from meeting notes, and this skill handles all the extraction and cross-referencing.
• file and note handle everything else you want to capture — articles, documents, quick thoughts.
• start-day and end-day create the daily rhythm that keeps your Work Board current.
• week-review is where compound knowledge happens — Claude looks across the whole week and surfaces patterns you might have missed.

---

Adding More Skills

Skills are just Markdown files in your .claude/skills/ directory. Adding a new skill is as simple as creating a new file.

Where skills live:

.claude/
└── skills/
    ├── process-meeting/
    │   ├── process-meeting.md    (the skill file)
    │   └── references/           (optional supporting files)
    ├── file.md
    ├── note.md
    ├── start-day.md
    ├── end-day.md
    └── week-review.md

Creating a custom skill:

1. Create a new .md file in .claude/skills/
2. Add frontmatter with name, description, and triggers
3. Write the instruction body
4. Claude picks it up in the next conversation

Example: A custom "prep for 1:1" skill

---
name: prep-1on1
description: Prepare for a 1:1 meeting by pulling together
  recent context about the person.
triggers:
  - "prep for 1:1"
  - "prepare for one on one"
  - "1:1 prep"
---

## Instructions

When preparing for a 1:1:

1. Load the person's People profile
2. Find their recent meeting notes (last 2 weeks)
3. Check Work Board for items they own or are waiting on
4. Check for any recent decisions they were involved in
5. Summarize:
   - What they have been working on
   - Open action items (theirs and yours)
   - Topics to discuss
   - Any concerns or positions to follow up on

Save the file, and next time you say "prep for my 1:1 with George," Claude knows exactly what to do.

Skills are shareable. Because they are plain Markdown files, you can share skills with colleagues, publish them as templates, or fork someone else's skill and customize it. There is no proprietary format — just text files with instructions.