What's new: agents you can mention, schedule and email

Mention an agent in a thread, run one on a schedule or start one with an email. Here are the biggest things we shipped in Markbase over the past month.

Markbase Team2 min read

It’s been a busy month here at Markbase, so we’ve rounded up the new features we think you’ll want to know about.

The biggest arrivals were Dispatch, which starts coding agents on your own machines, and team skills, which keep everyone’s agents on the same version of a skill through the mb CLI. Much of what follows builds on them.

More ways to start an agent

Mentions. Write @Markbase in a Linear issue, a GitHub issue or pull request, or a Slack thread. An agent picks up the work and answers in the same thread, so the conversation stays where your team is already having it.

Schedules. Run an agent every weekday morning or once a month, in your time zone. The prompt can live in a shared document, so you and your agents can improve it between runs.

Emails. Give a lane its own email address and forward it a customer’s bug report. The agent gets the message and its attachments, and you choose who’s allowed to send to it.

Due dates. Some work has to wait, like a feature in customer testing. Set a due date on the document that holds the plan, and when the day comes, Markbase reminds you or hands the work to an agent.

You stay in charge

Approvals. Some jobs should wait for a person. With approvals on, you see the exact prompt the agent will receive, and nothing reaches a runner until someone says yes.

Job limits. A burst of alerts no longer has to mean a burst of agents. Budgets cap how many jobs can start, and related alerts can fold into a single job.

Work register. Agents can claim what they’re working on and see what’s already taken, so two of them are less likely to chase the same problem.

See how the work went

Job reports. Agents now end a job by saying what they concluded: the work is done, there was nothing to do, or they need something to carry on. Markbase takes it from there, for example by holding back repeats of a false alarm.

Live activity. Turn this on for a profile and you can follow a Claude Code agent’s messages and tool calls from the job page while it runs, with no waiting until the end.

Profiles. A profile sets how a job runs: the agent tool, the model, the effort, the sandbox and the time limit. It can also say which runners may take the job, so a nightly tidy-up and an urgent bug don’t have to run the same way.

A few more we like

  • Import skills from a public Git repository into a skill set.
  • Temporary webhooks give your agent a short-lived URL, so it can see what a service really sends while wiring up an integration.
  • Git-aware docs mark a design that’s still on a branch as proposed, so agents on main don’t build on plans that might change.
  • The official MCP Registry now lists Markbase, which makes us easier to find from your MCP tools.

That’s the month! If you try any of these, we’d love to hear how it goes. New to Markbase? Come take a look or book a demo.