Pause one agent. Your pipeline keeps running.

When you publish seven posts in quick succession, you do not want a social media agent firing a draft for each one. You want to pause that step, run the batch, then bring it back.

That is what forge-agent makes possible.

Archive one job. Only that job stops.

An AgentJob in Smeldr has the same lifecycle as any other content: Draft, Published, Archived. Published means active. Archived means stopped. To pause an agent, you archive its job. To resume it, you publish it again.

No code change. No redeploy. No configuration to update. Nothing else in your pipeline changes, because the other agents were never connected to this one. They each subscribe to their own content type and their own signal. One goes quiet. The rest keep running.

What this looked like in practice

We published seven devlog posts in a single session. The social media agent is wired to fire a draft whenever a post is published. Without intervention, it would have generated fourteen social drafts automatically, one per post per platform, before we had reviewed any of them.

Instead: archive the agent job before the batch. Publish the posts. Review and confirm. Publish the agent job again when ready.

The publishing pipeline ran without interruption. The social drafts will come when we want them, not as a side effect of a batch operation.

The principle behind it

Each agent in Smeldr knows exactly one thing: which content type triggers it and what it produces. That scope is the ContentTypeFilter. It means every agent is independent. Adding one does not affect the others. Neither does removing one.

That independence is what makes the pause meaningful. When you archive an agent job, the boundary of that change is the agent itself. Everything outside it is unaffected.

*See Add a step. Change nothing else. for how agent composition works.* *See forge-agent for wiring and configuration.*