Magic LibraryConnectors: give Super Magic more reach
Out of the box Super Magic works with Thread’s own data and its built-in integrations. Connectors extend it to external tools over MCP — so Super Magic (and Thread MCP) can read and act in systems your desk already lives in. Add external tools to Super Magic with connectors (Super MCP) is the setup path; the high-value ones for automation work:Connector tools need their own sign-in and inherit the same permission model as every other Super Magic tool — they show up only once connected, and writes still run under your identity with confirmation.
Skills: package a workflow once, run it everywhere
A skill is a saved, prompt-first workflow — a proven way of doing a task that anyone can run in Super Magic by name. The Automation & Flows category is full of them, and the Connectors category shows real skills built on the connectors above. When you have a prompt that reliably does the job, don’t leave it in your head — make it a skill.1
Prove the prompt by hand
Run it in the Super Magic composer until it works every time. A skill is only as good as the prompt underneath it.
2
Turn it into a reusable skill
Use the Skill Authoring Coach to shape the prompt into a clean, named skill — clear inputs, defined output, no dependence on one person’s context. Keep the instructions under 3,000 characters; the coach counts them for you and shows you where to cut.
3
Wire it to a Flow if it should run itself
A Flow that fires Super Magic on a ticket event is a Super Magic Agent (Super Magic + Flows). The Flow carries its own prompt and a write scope — it doesn’t call your saved skill, so paste the proven prompt in and give it only the write tools you’re comfortable with it using unattended.
The 3,000-character limit
Skill instructions and Flow agent prompts are capped at 3,000 characters. A skill saved before the cap keeps running, but it can’t be saved again until it’s shortened — so the limit shows up the moment you go to edit a long one. Almost every prompt that’s over the line is over because it explains why a step matters two or three times. Say it once, in the imperative, and most drafts land well under. When that isn’t enough, name a base skill instead of restating a contract you share with other skills — PSA Note Discipline for plain-text notes, Write Guardrails for confirmation gates, Connector Degradation for a missing integration, Sweep Honesty for result caps.Automagically: runbooks that execute unattended
For multi-step operational work — the runbooks a technician would otherwise click through by hand — Automagically runs the procedure for you. Automagically getting started is the on-ramp: define a runbook once, and it executes the steps unattended, with the same read-free / write-confirmed discipline the rest of Thread’s automation follows. Reach for Automagically when the work is a procedure (a sequence of steps that’s the same every time) rather than a decision (route this / respond to that). Pair it with your Flows and Intents: the Flow decides when, the Intent handles the conversation, and Automagically runs the steps.Where the pieces fit
You now have the full automation toolkit. A quick map of which layer owns what:- Flows / status automations — deterministic routing and state, no code.
- Intents — conversational handling of a request type by the Triage Agent.
- Super Magic skills — packaged, reusable workflows a person runs by name in Super Magic.
- Connectors — extend that reach to external tools over MCP.
- Thread MCP / API / Rewst — the same tool layer in your own client or code.
- Automagically — unattended execution of repeatable runbooks.
Next
That’s the course. Take a pattern from your own desk and ship it — an Intent, a skill, or a runbook — then come back and browse the full library for the next one.Back to your course
The hub: your starter kit, build ritual, and FAQ.
Automation & Flows skills
The full Skill Library category built for your role.