add-skill

Add-Skill — The Bridge to Personal Skills

You help the user build a new skill that's specific to their business. Cowork AI OS ships universal skills. This is how the user makes Cowork AI OS personal.

You don't build the skill yourself. You write a structured spec, then hand off to Cowork's built-in Skill Creator, which is purpose-built for skill construction. Cowork AI OS is the orchestrator. Skill Creator is the carpenter.

Pre-flight check (do this first, every time)

Before anything else, verify Skill Creator is enabled. If it's not, /add-skill cannot work — Skill Creator is what actually writes the skill file.

Step 1: ask the user:

"Quick check before we start: open Cowork → Settings → Skills. Is Skill Creator toggled ON?"

Wait for yes or no.

If yes → proceed to Step 1 of the workflow below.

If no → walk them through enabling it:

"We need to turn it on first.

1. Open Cowork → Settings → Skills.
2. Find Skill Creator in the list.
3. Click the toggle to ON.
4. Tell me when it's on.

If you don't see Skill Creator in the list, your Cowork might be out of date. Check claude.com/download for the latest version."

Wait for confirmation. Then continue.

Read these first

1. about-me/about-me.md, business-brain.md, connections.md — to ground the skill in their actual business
2. about-me/writing-rules.md — if the skill produces any text, voice rules apply
3. the 4 Cs in the appendix below - the new skill should improve at least one C

Workflow

Step 1 — Clarify the skill (4 questions max)

Ask these one at a time. Don't ask more than 4. Default what you can.

1. "What's the trigger? When should I run this skill?" (e.g. "every Monday morning," "when I say 'lead recap,'" "after a client books a call")
2. "What does it read or look at?" (files, connectors, the user's input — be specific)
3. "What does it produce?" (a draft, a summary, a file in a specific folder, an action like a Gmail draft)
4. "What's the success test?" (how will the user know if the skill is working well — e.g. "the recap saves me 20 min on Monday morning")

If the answers are crystal clear after 2 questions, stop asking and move on.

Step 2 — Draft the spec

Format a skill spec the user can review:

## Proposed skill: ~~name

**Trigger:** [what fires it — slash command + plain English alias]

**Reads:**
- [file 1]
- [connector 1]
- [user input]

**Does:**
1. [step 1]
2. [step 2]
3. [step 3]

**Writes:**
- [output file or action]

**Voice/style rules:** [if it produces text — e.g. "follow writing-rules.md, run anti-AI kill list scan"]

**Why this matters:** [one-sentence leverage explanation]

Show the spec. Say:

"That's the skill. Ready for me to hand it off to Cowork's Skill Creator to actually build the file? Type build it or tell me what to change in the spec."

Wait. Iterate the spec until they say build it.

Step 3 — Hand off to Skill Creator

(Skill Creator is already enabled — the pre-flight check at the top of this skill confirmed that.)

Tell the user:

"Handing this spec off to Skill Creator now. Here's what to do:

1. Open a new Cowork task.
2. Paste the spec I'm about to give you plus this instruction:

'Use Skill Creator to build a custom skill from this spec. Save it as a project skill in my workspace so it loads in every Cowork session.'"

Then provide the spec (the markdown block from Step 2) ready to paste.

Step 4 — Confirm and test

After they've run the Skill Creator handoff, ask:

"Did Skill Creator finish? Type yes — what next when it does."

When they confirm:

1. Help them test the new skill by triggering it.
2. If it doesn't work right, walk them through editing it.
3. Once it works, append to memory.md:

### YYYY-MM-DD — Built new skill: ~~name
- **Trigger:** [trigger]
- **Spec source:** /add-skill conversation on [date]
- **Test result:** [pass / pass with edits / failed]
- **What it unlocks:** [one-sentence leverage win]

Step 5 — Update audit-log

The new skill bumps Capabilities. Append a note to about-me/audit-log.md:

### YYYY-MM-DD — Custom skill added: ~~name
- **C affected:** Capabilities
- **Before:** [score]
- **After:** [score]

This makes the next /audit and /tune-up reflect the gain.

Edge cases

"I want a skill that does 5 different things"

Politely refuse. Skills should do one thing well. Tell the user: "Let's build the most leveraged of those 5 first. The other 4 can become their own skills later, and they'll be easier to write because each is focused."

"I want a skill that exists in a plugin already"

Check the recommended plugins appendix below. If Anthropic's productivity / sales / marketing / etc. plugin already has it, recommend installing that plugin instead. Custom skills are for things no existing plugin covers.

"I want a skill that needs a tool I haven't connected"

Tell the user: "First we need to connect [tool]. Want to walk through that now, or pick a different skill to build first?"

"I changed my mind, I don't want to build it"

No problem. Append to memory.md:

### YYYY-MM-DD — Considered new skill: ~~name (not built)
- **Reason:** [user's words]
- **Spec saved at:** [path, if you saved it]

Tone

This is the moment Cowork AI OS becomes personal to the user's business. Treat it like the unlock it is. Don't over-celebrate, but don't make it feel routine either. Each new skill is the OS getting smarter about their world specifically.

---

Appendix: The 4 Cs framework (the audit rubric)

Score the user Cowork AI OS on a 0-100 scale. 25 points per C.

C1 - Context (0-25)

How much does Claude know about the user and their business?

• 0-8: No about-me/ files, or files have ~~placeholders unfilled.
• 9-16: About-me/ has the basics but writing-rules.md has fewer than 3 samples.
• 17-22: All 4 about-me/ files complete. Writing samples real. Memory.md has 5+ entries.
• 23-25: Above plus business-brain.md kept current and memory.md gets appended every session.

C2 - Connections (0-25)

How many of the 7 buckets (Revenue, Customer, Calendar, Comms, Tasks, Meetings, Knowledge) are reachable?

• 0-8: 0-2 buckets reachable.
• 9-16: 3-4 buckets reachable.
• 17-22: 5-6 buckets reachable.
• 23-25: All 7 reachable with safe permissions (writes need approval).

C3 - Capabilities (0-25)

How many things can Claude actually do?

• 0-8: Cowork AI OS only, never run a skill beyond /onboard.
• 9-16: Cowork AI OS skills used regularly. No additional plugins.
• 17-22: Cowork AI OS plus 1-2 Anthropic role plugins. 1-2 custom skills built via /add-skill.
• 23-25: Above plus custom skills get refined every few weeks.

C4 - Cadence (0-25)

How much works on a schedule, vs only when asked?

• 0-8: No scheduled tasks.
• 9-16: Morning brief scheduled. Nothing else.
• 17-22: Morning brief plus weekly tune-up plus at least one custom routine.
• 23-25: Multiple routines firing reliably. Computer-awake reliability solved.

---

Appendix: Recommended next plugins (after Cowork AI OS)

For most users:

• productivity - manage tasks, calendar, daily workflows. Most universal pick.

By role:

• Coach, agency owner, creator: marketing plugin
• Founder, solopreneur: productivity plugin
• Sales-led business: sales plugin
• Client services: customer-support plugin
• Product or SaaS: product-management plugin
• Designer: design plugin
• Legal, finance, data: legal, finance, data plugins
• Multi-tool searcher: enterprise-search plugin

Install via Cowork: Customize -> Browse plugins -> find the plugin -> Install.

When to build custom via /add-skill instead: when the user workflow is specific to their business (their client onboarding, their lead qualification, their weekly internal report) and no plugin covers it.