glossary · Running it

Agent catalog

definition

An agent catalog is one shared, versioned list of a team's AI agent definitions, so every person and every run uses the same named agent at a known revision instead of a private copy of its prompt.

in one line: One versioned list of your agents, named on every step.

in context

Where does an agent catalog fit?

One catalog, every revision kept. Every save is a numbered revision. One is active. Roll back by moving the pointer.

Revisionsnumbered, never changed

Step agentsnamed on the step

In usedisable is refused

What the catalog keeps for each agent.

two ways to say it

What is an agent catalog, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

Agent files drift between repositories and laptops. A catalog keeps one copy, with its history.

Each workflow step names the agent it needs, and gets its active revision.

for engineers

agents_manage: create, update (a new revision), set_active, disable. No delete.

The body needs ## Prerequisites, ## Role and ## Rules. The model names a workspace tier.

the longer answer

Why does an agent catalog matter?

AI coding tools let you define subagents: a name, a description of when to use it, a model and a prompt. Claude Code reads them from `.claude/agents`, OpenCode from `.opencode/agents`. The files are quick to write and quick to copy, and that is the problem. Each repository and each laptop ends up with its own version, and nobody can say which text ran last night.

A catalog moves the definitions to one place the whole team reads. Each change is saved as a new revision instead of overwriting the old one, so there is a history to compare and a known version to go back to. A run then names the agent it needs, and the catalog answers with the exact text.

in convops

How does an agent catalog work in ConvOps?

In ConvOps, the agent catalog belongs to the workspace. An agent has a description, a model tier and a body with Prerequisites, Role and Rules sections. Every save writes a numbered, immutable revision, and one revision is active; setting an older one active is a rollback that deletes nothing. Agents are disabled, never deleted, and a disable is refused while an enabled workflow step names the agent. Teams group agents, and each workflow step names the agent that does it. The convops command line imports and exports `.claude/agents` and `.opencode/agents` files both ways.

questions

What do people ask about an agent catalog?

How is an agent catalog different from a .claude/agents folder?

A folder holds one copy per repository, edited in place. A catalog holds one shared copy per workspace, and every change is a new numbered revision, so the history stays and any revision can be made active again.

Can I keep my existing agent files?

Yes. The convops command line imports .claude/agents and .opencode/agents into the ConvOps catalog and exports the catalog back as files, with a preview that writes nothing in both directions.

What does an agent catalog connect to?

Agent teamAgents grouped for one job, pinned or following.Model tierAgents ask for a level; the workspace picks the model.Agent executorWhich agent runs the work, and where.Agent instruction file (CLAUDE.md, AGENTS.md)CLAUDE.md or AGENTS.md: what every session reads first.

Agents in ConvOps Agent team Model tier Agent instruction file

your next step

Teach it once. Then let it run.

on your ownStart freeConnect your AI tool in one line. Nothing to install.with your teamBook a demoWe show ConvOps on your team's real work.keep readingAgent teamAgents grouped for one job, pinned or following.