What it’s for
Telling people what an agent found, where they already are. A workflow’s last step posts a run summary to a channel; a chat agent replies in the thread where someone asked about a finding; a triage agent reads the last day of#security to see what’s already been discussed.
How agents appear in Slack
Slack grants Amplify a user token for the bot account, not a bot-app token. That has two consequences:- Messages and reactions appear as the bot account — its name and avatar — not as an app.
- Agents can only read channels and conversations the bot account is a member of. Invite it to a channel and agents can read and post there; leave it out and that channel doesn’t exist as far as agents are concerned.
Before you connect
- Create the bot Slack account in your workspace and set its display name and avatar. See the bot account.
- Invite it to every channel agents should read or post in. It sees nothing else.
- If your workspace requires admin approval for apps, the connect flow pauses on a Slack approval request.
A Slack admin can approve it from Admin → Apps → Requests to install, or pre-approve ahead of time
with an app-approval rule for app ID
A0C28V9SYJK(Amplify Console). Amplify Console isn’t listed in the Slack Marketplace, so there is no Marketplace page to send them. - Sign your browser into the bot account, then follow the connect flow.
What agents can do
Agents are instructed to post only where they were asked, to reply in a finding’s existing thread rather
than starting a new top-level message, to end each post with a note that it was generated by an Amplify
Console agent, and never to direct-message anyone who didn’t ask. History reads are for conversations you
ask about, not for sweeping channels.
What agents can’t do
- Delete or edit messages, or remove reactions
- Create channels, invite or remove members, or change any workspace or channel setting
- Read a channel the bot account isn’t in
- Post as an app, with a custom name or avatar per message
- Receive events or respond to mentions on their own — agents act only inside a run or chat turn
