For a while, every task in The Agency carried two things at once: what you wanted done and how to do it. If you had three tasks that all started with "triage my inbox," you'd written the procedure for triaging your inbox three times. And when you wanted to tweak how it worked, you had to remember to update all three.

That always felt off. The methodology of "how to triage email" doesn't belong to a specific task. It's expertise. It should live somewhere your agent can reach for it whenever the situation calls for it — in a scheduled morning briefing, in an ad-hoc chat, in a webhook that fires when an order comes in.

So we split it apart. We're introducing Skills, and the old Tasks are now Workflows.

Recipes and ingredients

The mental model is simple:

Skills are the ingredients. Atomic procedures. "How to triage email." "How to write a research report." "How to prep for a meeting." Each one stands alone.
Workflows are the recipes. They describe what you want — what to produce, when to run, what inputs to use — and they call on skills to do the work.

Here's a real example. The Morning Briefing workflow looks like this:

Sample prompt

Produce a tight briefing I can read in 2 minutes. Lead with what's urgent. Run @email-triage on my work inbox for the last 24h. Then use @calendar-review for today, flagging conflicts or prep needed. Finish with @research-topic on {{focus}}.

Three @-mentions, one {{variable}}. When the workflow runs, your agent pulls in the full procedure for each skill, fills in your inputs, and gets to work. You're describing the recipe at a high level. The skills handle the details.

Edit email-triage once, and every workflow that calls it — your Morning Briefing, your Friday Wrap, your customer-support catch-up — gets the update for free.

An open format

Skills follow the open Agent Skills format pioneered by Anthropic. It's plain markdown with frontmatter, the same shape used in Claude Code, Cursor, and other tools:

---
name: email-triage
description: Process inbox by priority, draft replies, flag follow-ups
when_to_use: |
  When the handler wants to catch up on email or clear unread.
inputs:
  - name: scope
    default: "unread last 24h"
---

# Email Triage

Goal: leave the handler with a clear picture of what needs action.

## Procedure
1. Pull threads in {{scope}} using GMAIL_FETCH_MESSAGES
2. Categorize: needs_reply / needs_action / fyi / archive
3. For obvious replies, draft via GMAIL_CREATE_EMAIL_DRAFT (never send)
4. Produce a report grouped by category

That's the whole format. Because it's open, you don't have to write all your skills in our editor. Find a SKILL.md you like in a GitHub repo, paste the URL into Settings → Skills, preview it, and approve. Your agent now has a new piece of expertise it can reach for.

We've started with a set of system skills out of the box — email-triage, calendar-review, meeting-prep, research-topic, summarize-pdf, send-update, price-watch, and a handful more. The workflow templates in the library are already wired to use them, with the Morning Brief and Weekly Review delivered as workflow compositions that call the atomic skills.

It works in chat, too

The same library powers chat. When you type "help me triage my inbox," your agent doesn't re-invent the procedure. It sees email-triage in your enabled skills, loads it, and follows the same recipe a scheduled workflow would. You'll see a small Using: email-triage indicator in the chat so you know what your agent reached for.

That's the part we're most excited about. The same expertise drives everything — scheduled runs, webhook triggers, ad-hoc chat, voice messages on Telegram. One skill, every surface.

Why it matters

A few practical things change.

Your library compounds. Every skill you add — whether you write it, import it from GitHub, or get it as part of a future update — makes your agent sharper. It's a one-time addition that pays off across every workflow and every chat from then on.

You can share procedures. A SKILL.md is a markdown file. Drop it in a gist, commit it to a repo, paste it in a Slack message. Anyone running The Agency (or Claude Code, or any compatible tool) can import it.

Workflows get easier to write. You're describing intent — "run an email triage, then a calendar review" — not re-deriving the procedure each time. The recipe is short. The ingredients are reusable.

What's new in the app

A few things to look for:

Workflows has its own spot in the top nav, replacing Tasks. Existing tasks were migrated automatically — same schedules, same triggers, same behavior.
Settings → Skills is where the library lives. List, toggle, view, import, or write your own.
The workflow editor now supports @skill-name autocomplete and {{variable}} inputs. Type @ to pick a skill, type {{, and your agent will collect the value when the workflow runs.
Tasks still exist, but they're now the deferred-action thing they always wanted to be — "remind me Wednesday at 3pm to call Mom." One-off, time-bound, no procedure required.

Try it

If you've used The Agency for a while, your existing tasks are now workflows. They'll keep running on the same schedules. The next time you open one to edit, try replacing a paragraph with a skill reference and see what changes.

If you're starting fresh, open the workflow library, pick something close to what you want, and notice the @-mentions in the content. Those are your starting ingredients.

Either way, head to Settings → Skills and browse what's there. The library is small today, but it's the foundation everything else now sits on — and it grows every time you add another skill.