Problem

Scheduled tasks are a powerful part of Humr, but today they can only be set up by a human through the UI. This means an agent can't follow up on its own work.

Concretely: if an agent is working on something and decides mid-session that it would be useful to check back on it later — monitor a site daily, run a weekly report, retry something after a cooldown — it has no way to do that. It can only ask the human to go create a schedule, which defeats the point of giving the agent autonomy.

We want agents to be able to schedule themselves.

Proposed solution

Give agents a way to register a scheduled task directly, without human intervention.

From the agent's perspective, it should be able to say "run this prompt on this cron schedule" and have the platform handle the rest. The schedule it creates should behave identically to one a human would create through the UI.

Scope

An agent should only be able to manage schedules for itself — it cannot schedule work on behalf of other agents or instances.

Beyond creating schedules, agents likely need to list, update, disable, and delete the schedules they've made, otherwise they can't undo their own decisions.

Transparency

Schedules created by an agent should be distinguishable in the UI from ones created by a human, so the human owner can see at a glance what their agent has scheduled itself to do. This is important for trust — the human should never be surprised by work their agent scheduled.

Out of scope for this issue

• Letting agents schedule work on instances other than their own.
• Rate limits or quotas on how many schedules an agent can create (worth discussing, not blocking).