Skills are the new prompts, content designers

🎁 The folder beats the stack. → Take the free starter folder my workshop students get

Folder structure (C-core / R-references / O-output / M-memory), agent skill files, Claude Code config, Obsidian-ready. No subscription. No vector DB. No SaaS. Just a folder.

---

Last month a thing happened.

Anthropic moved Agent Skills from "feature" to "open standard with a partner directory." Atlassian, Canva, Cloudflare, Figma, Notion, Ramp, Sentry — all on the launch page. Then on May 5 they shipped 10 finance agent templates as proof.

If you write content for a living, this is your month.

Here's why.

TL;DR

→ The Gift: Free starter folder — the exact architecture this issue is about, ready to clone tonight.

→ The Signal: Anthropic made Skills an open standard — and the directory looks a lot like a content designer's job description.

→ The Build: 10 finance agent templates shipped as reference architectures. Skills + connectors + subagents. Not a demo.

→ The Pattern: Skills are the unit. Markdown is the format. Three sections is the spec. Try it tonight.

🔥 The Signal of the Month — Skills became the standard

A Skill is a folder of markdown an agent reads instead of improvising. What this is for. How to do it. How to know it's done. Three sections. That's the whole spec.

Last month Anthropic made it official. The Agent Skills directory launched with seven partners on day one — Atlassian, Canva, Cloudflare, Figma, Notion, Ramp, Sentry. The same playbook they ran with MCP twelve months ago: ship a spec, line up partners, watch the ecosystem follow.

Anthropic's own engineering team wrote: "Context engineering is the load-bearing skill of 2026." That's not a quote about prompt tricks. That's a quote about documentation that runs.

→ Anthropic engineering: Equipping agents for the real world with Agent Skills

Here's what content designers keep missing.

The thing the agent reads to behave like your brand IS content. Tone of voice → Skill. Editorial guidelines → Skill. Onboarding flow → Skill. Refund policy → Skill. The voice file you wrote three years ago that "nobody uses?" That's not a voice file. That's a Skill that hasn't been wired up yet.

The people who already know how to write structured, reusable, testable instructions for humans — content designers, UX writers, technical writers, instructional designers — just inherited the operating manual for agents. Same craft. Different reader. The reader is a model now.

Skills aren't prompts. They're documentation that runs.

→ The New Stack: Agent Skills, Anthropic's Next Bid to Define AI Standards

🛠️ Build of the Month — Anthropic's 10 finance agent templates

May 5: Anthropic shipped ten ready-to-run agent templates for financial services. Pitch builder. Meeting preparer. Earnings reviewer. Model builder. Market researcher. Valuation reviewer. GL reconciler. Month-end closer. Statement auditor. KYC screener.

Each template has the same anatomy: Skills + connectors + subagents. Skills are the instructions. Connectors are governed access to FactSet, S&P Capital IQ, Morningstar, the firm's own warehouses. Subagents are smaller Claudes called for specific sub-tasks like comparables selection. They drop into Claude Cowork, Claude Code, or the Claude Platform as Managed Agents.

This is the first public industry-specific agentic content design playbook. Vertical SaaS just got a new bar. If finance has 10 reference Skills, healthcare gets 10. Legal gets 10. Retail gets 10. Travel gets 10.

The content designers who write those Skills set the standard their entire industry runs on.

→ Anthropic: Agents for financial services

⚡ The Stack — 6 things that shipped this month

→ Claude Design (Apr 17) — text → prototype, slides, landing page, full site. You can ship a working brand artifact without a designer for the first round.

→ Claude in Microsoft 365 (May 5) — Excel, PowerPoint, Word, Outlook get Claude in-app. Where your stakeholders already live. Content review without context-switching.

→ Mastra — TypeScript agent framework. Replit's Agent 3 went 80% → 96% task success after switching. Marsh McLennan deployed it to 75,000 employees. TS-first means content people without Python can finally read the agent code.

→ Figma's use_figma MCP — agents read the design system as source of truth, push tokens back to code. (MCP — basically a USB port for AI agents to talk to your tools.) Design system + content together = the agent's brand brain.

→ HubSpot Breeze Brand Voice — voice profile applies across blogs, emails, pages, SMS, social. First mainstream "voice file as a Skill" in a tool you already pay for.

→ OpenAI Workspace Agents (Apr 22) — shared cloud agents in ChatGPT. Plain English defines the workflow. Your CEO will start asking why Marketing doesn't have one yet.

🧠 The Pattern — Skills are the unit. Markdown is the format.

Stop writing prompts. Start writing Skills.

A Skill is a markdown file in a folder your agent reads. Three sections:

1. What this is for — one paragraph
2. How to do it — the steps
3. How to know it's done — the success criteria

That's it. No framework. No SDK. No vendor lock-in.

Why this works now: Anthropic's 2026 Agentic Coding Trends Report calls context engineering "the load-bearing skill of 2026." A 754-Skill cybersecurity library shipped on GitHub last month, mapped to NIST + MITRE. The directory format works because it's the same pattern that already works for humans: docs.

Try this week: Open a folder called /skills. Pick the ONE workflow you'd hand off to a teammate if you trusted them. Write it as <workflow-name>.md with those three sections. Point your agent at the folder. Run it once. Watch what breaks. Fix the Skill, not the prompt.

→ Reference: 754-Skill cybersecurity library on GitHub — what shape + structure looks like at scale.

→ Or skip the blank folder: take the starter template I give my workshop students (free) — C-core / R-references / O-output / M-memory, agent skill files, the CLAUDE.md that wires it together, Obsidian-ready. No subscription. No vector DB. No SaaS. Just a folder.

🔗 Quick Links — the rest of the month

→ Smashing Magazine: Designing for Agentic AI — practical UX patterns — three patterns every agent UI needs: Pre-Action consent, In-Action transparency, Post-Action safety.

→ Smashing Magazine: When 'Production-Ready' Becomes A Design Deliverable — Carrie Webster on what's gained and lost when the line between design and engineering disappears.

→ Simon Willison: Vibe coding and agentic engineering are getting closer than I'd like — the framing piece on what "agentic" actually means now.

→ NN/g: Less Chat, More Answer — site AI chatbots need to get to the point — direct, scannable, expandable answers. The new content design rule for AI surfaces.

→ Ruben Hassid: Stop treating Claude like a search engine — three production agents (Email + Brief + Content) saving 2 hours/day.

→ Ioana Teleanu: youhealth.me, prompt-to-MVP in 6 hours — what design system + content + code can do when they're one workflow.

→ Anthropic 2026 Agentic Coding Trends Report (PDF) — the full quote in context: "context engineering is the load-bearing skill of 2026."

💬 The Prompt — answer me one thing

What's the ONE workflow you'd turn into a Skill this month, knowing it's just a markdown file your agent reads?

Reply to the next issue and tell me. I'll feature the best three.

---

If this was useful, forward it to one person who's still writing prompts.

— Yuval

P.S. The folder beats the stack. A head of design showed me his AI tool spreadsheet last week — six-figure tool budget in 90 days, zero shipped. Then I opened mine. Markdown files. Obsidian. Agents reading and writing them. Same architecture Figma just adopted for their canvas agents. → Take the same starter folder I give my workshop students (free, no subscription).