Create custom agents
Every team has a short list of jobs that aren’t hard, just relentless. Filing the same report every Friday. Summarising the weekly product video so people who skipped the meeting stay current. Scheduling a call when a deal moves to the next stage. Reading a busy support channel and flagging the three messages that actually need a human. None of it is interesting work, and all of it has to happen anyway.
A custom agent is how you stop doing it by hand. You’re not writing a script or wiring up a flowchart — you’re describing the outcome you want, the way you’d brief a new teammate, and letting the agent figure out the steps.
Describe the job, not the code
The unit of a custom agent is a workflow you already do. If you can explain it to a colleague in a few sentences, you can build it. Skynet turns that description into an agent that plans the work, calls the tools it needs, and uses your connected memory so it already knows the context — your Slack threads, your docs, your tickets, your PRs.
Here’s the kind of brief that’s enough to get started:
Notice what that brief contains: a trigger (Friday at 4pm), the sources to read (a Slack channel, linked PRs), the work to do (summarise, pull, flag), and where the result goes (back to the channel). That’s the whole shape of an agent. You don’t specify how to read Slack or how to fetch a PR — the agent handles that.
Build one in five steps
Describe the workflow
Open a new agent and write the job in plain language — what should happen, what it reads, and what it produces. Be specific about the output you want. “Summarise the channel” is vague; “five bullets, under 150 words, flag blockers” is something an agent can actually deliver.
Connect tools and memory
Give the agent the integrations it needs — Slack, GitHub, Notion, and more. These aren’t read-only: the agent can post, comment, and update where you allow it. Point it at the right memory sources so it carries the context a teammate would, instead of starting cold every run.
Set a trigger
Decide when it runs. Leave it on demand for jobs you kick off yourself, or put it on a schedule — every weekday morning, every Friday afternoon, the first of the month. Scheduled agents run whether or not you’re online.
Test it
Run it once and read what it produced. If the summary missed a thread or the tone is off, adjust the brief and run again. This is the part worth not skipping — a few test runs now save you from a wrong report going out automatically later.
Share it with your team
Once it’s solid, share the agent so anyone on your team can run it or rely on its schedule. One person builds the triage agent; the whole team stops missing tickets.
What you get back
The point isn’t novelty — it’s the hour you reclaim every week, and the small things that stop slipping. A scheduled agent files the report before you’ve had coffee. A triage agent catches the urgent ticket while the channel is still loud. And because each agent draws on your shared memory and can act in your tools, the output reads like a teammate did it, not a macro.
Start with one annoying task. Build the agent. Then go find the next one.
Frequently asked questions
No. You describe the job in plain language — what it should do, what it reads, and what it produces. Skynet handles the rest. If you can brief a colleague, you can build an agent.
Yes. Set an agent to run on demand when you kick it off, or on a schedule — daily, weekly, monthly, whatever the job needs. Scheduled agents run on their own, whether or not you are online.
Yes. Once an agent works the way you want, share it so teammates can run it themselves or rely on its schedule. Build it once, and the whole team benefits.
Agents work with your connected integrations — Slack, GitHub, Notion, and more — where they can take action, not just read. They also draw on your unified memory, so they already know the context from your sources before the first run.