A Slack message when your AI needs a decision

What actually arrives

One message, built to be judged at a glance. It carries the headline reason the work stopped, and it names the agent that stopped, so a channel carrying several pieces of work at once stays readable. Under it sits the single load‑bearing sentence, the one thing you need in order to know whether this is yours to answer now or later.

What it does not carry is the argument. The full walk stays behind the link, because a notification that tries to contain the reasoning is a notification that ages badly: it is written once and read later, and by then the thing it describes has moved. Everything long is on the other side of the link.

The link is an address, not a copy

The link in the message does not carry the content of the item. It carries the item’s address. When you open it, your own AI reaches back and pulls the item with its reasoning, live, at the moment you are reading, which is the only moment the answer needs to be current in.

That is also why the link is safe to put in a channel. It is opaque and it expires: it carries no credential, and it reveals nothing about the item or the workspace to anyone holding it who is not signed in. An expired link, a valid link and a forged one all land in exactly the same place.

Which moments land there is your switch

Every pasted endpoint carries its own routing rules: which classes of moment are allowed to land on it. A question that needs answering and a run that finished are not the same event and do not deserve the same channel, so they are separate switches rather than one on‑or‑off for everything.

The classes Unl raises about itself, rather than about your work, are off until you turn them on. That is the default in the code and not a setting to be discovered later: an alerting product whose defaults fill your channel has taught you to mute it by the end of the first week.

What is deliberately not decided for you

Timeouts, escalation rules, what counts as blocking, cadence, and whether work holds or proceeds while it waits: none of these are set here, and none of them are inferred. A knock is an emission, not a wait. The work does not block because a message was sent, and nothing changes state because you did not read it.

That boundary is the point rather than an omission. The moment a tool starts deciding what is urgent for you, it has started deciding, and the reason to be told at all was to keep that judgment where it belongs.

Getting it running

Create an incoming webhook in the Slack workspace and channel you want, and paste the URL into Unl. There is no app to install on our side, no bot to add, and no scopes to review, because the endpoint you pasted is the whole integration.

The URL is then held as a credential: encrypted at rest, shown back to you only as its last four characters, revocable whenever you want, and re‑pasting a fresh URL rotates it in place rather than leaving a second live endpoint behind. Slack’s own per‑message limit is respected, and where a message must be shortened to fit, the text shrinks and the link is never truncated and never dropped.

Paste a Slack webhook and the moments that genuinely need you arrive in the channel you already watch, carrying why the work stopped, which agent stopped it, and a link that opens the item in your own AI with the reasoning pulled live rather than copied into the message.

Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. The reach lane is live: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.

Read further

• The same thing, in Discord
• Paste any webhook, not just these two
• Stop babysitting the agent