The skill catalog: installable procedural memory for your agent

The agent's memory has two procedural layers. The first is the skills the agent learns by watching you (covered in Skills as procedural memory). The second is skills you install from a catalog: pre-built recipes maintained by us, by partners, or by the open developer community. This second layer is the one that scales the agent past the limits of what one user can teach it.

This post is how the catalog works, what's in it today, and how to write your own skill if the catalog is missing something you need.

What a catalog skill looks like

A skill is a small JSON-and-markdown record describing how to handle a specific kind of task. The metadata: name, description, version, maintainer, required tools, optional tools, sample triggers. The body: the recipe in natural language plus any deterministic logic.

A representative skill from the catalog: "schedule a recurring weekly check-in with my team."

name: weekly-team-checkin
description: |
  Schedule a recurring weekly check-in meeting. Picks a default
  time based on attendees' calendars. Adds a meeting agenda
  template. Sends the invite. Reschedules conflicts.
version: 1.4.0
maintainer: ajeris-team
required_tools: [calendar, email]
optional_tools: [slack]
sample_triggers:
  - "set up a weekly check-in with marketing"
  - "weekly sync with the engineering team"
  - "recurring 1:1 with my manager"

When you install the skill ("install the weekly-team-checkin skill from the catalog"), it becomes part of your agent's procedural memory. The next time you trigger it, the agent uses the recipe.

What's in the catalog today

The early catalog has ~80 skills across eight categories.

The catalog is curated. Skills go through review before publishing; we check for safety (no destructive defaults), quality (the recipe actually works), and brand (no spam). Most skills are written by the Ajeris team; partners and the developer community publish a growing share.

Industry packs, by request only

Important note for users coming from other AI products: the catalog includes industry packs (HVAC, salons, restaurants, lawyers) but they are opt-in. We never default-attach an industry pack based on user signals. The reason: most users are not running a service business. A photographer asking the agent to "book a meeting with Sara" should not get a contractor-flavoured booking flow.

You install an industry pack explicitly when you know you want it. From the manage page or by texting "install the HVAC pack."

Composition: catalog skills + personal skills

The two layers compose at runtime. When you ask the agent to do something, it considers:

1. Personal skills you have taught it (highest priority, these are your specific way).
2. Installed catalog skills that match the trigger.
3. Default agent behaviour.

If a catalog skill says "use a 30-minute slot for check-ins" and your personal skill says "use 25-minute slots so I have time to write the recap," the personal skill wins. The catalog provides defaults; you customize.

You can also patch a catalog skill in plain English without it affecting other users: "for the weekly-team-checkin skill, always include 'agenda from last week' in the meeting description." The patch is applied to your personalized version.

Publishing your own skill

Three paths.

Personal-use only. Just teach the agent. The skill stays personal.

Share with your team. Publish to your team workspace. Other users in the workspace can install. This is the model most companies use; it lets the team standardize "how we book customer-success kickoff meetings."

Publish to the public catalog. A small review process. The skill becomes available to all Ajeris users. We pay a small revenue share to skill authors whose work gets adopted broadly; details are in the developer docs.

The publishing flow is text-driven. You ask the agent to "publish the weekly-team-checkin skill to my workspace"; the agent stages it with metadata, asks you to review, and publishes.

You: publish my "send weekly-recap-from-meeting" skill to my workspace
 
Ajeris: Staged for publish:
        Name: send-weekly-recap-from-meeting
        Description: Send a weekly recap based on this week's
                     meeting transcripts to the team Slack.
        Required tools: meetings, slack
        Version: 0.1.0
 
        Review the recipe? <link>
 
You: looks good, ship it.
 
Ajeris: Published. Available to your workspace under
        Skills > workspace > send-weekly-recap-from-meeting.

What this enables

Three things at scale.

Faster onboarding for new users. A new Ajeris user does not have to teach the agent everything. They install the relevant pack ("startup founder pack" for example) and get 10-15 skills ready to go.

Team standardization without scripting. Companies that want every employee's agent to handle "schedule a weekly customer call" the same way publish one skill, install it across the workspace, and the standardization is shipped without any code.

Vertical depth without bloat. The agent's core stays generic. Industry-specific behaviour lives in opt-in packs. Photographers and HVAC companies get the same agent core; their specific workflows come from the relevant pack.

Try it

If you have ever thought "I wish my agent did X consistently," check the catalog first. If it's there, install. If not, write it and publish.