| name | slack-web |
| description | Slack web automation via agent-browser. Use when the user asks to check Slack, read messages, send messages, reply in threads, search Slack, check notifications, react to messages, send DMs, or any Slack communication task. Triggers include "check my Slack", "send a message on Slack", "reply to that Slack thread", "search Slack for...", "what's new on Slack", "read my DMs", "react to that message", "check Slack notifications", "message [person] on Slack".
|
| allowed-tools | Bash(agent-browser:*) |
Slack Web Automation
Interact with Slack through its web interface using agent-browser.
Core Workflow
Every Slack interaction follows this pattern:
- Restore session:
agent-browser state load ~/.slack-auth-state.json
- Open Slack:
agent-browser --headed open "https://app.slack.com"
- Snapshot:
agent-browser snapshot -i -c (get refs)
- Interact: Use refs to click, type, navigate
- Re-snapshot: After every action — Slack's DOM changes constantly
agent-browser state load ~/.slack-auth-state.json
agent-browser --headed open "https://app.slack.com"
agent-browser wait --load networkidle
agent-browser set viewport 1920 1080
agent-browser get url
agent-browser press "Meta+k"
agent-browser wait 500
agent-browser snapshot -i
agent-browser fill @eXX "general"
agent-browser wait 1000
agent-browser snapshot -i
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -c
Authentication
First-Time Login (Manual)
agent-browser --headed open "https://app.slack.com"
agent-browser get url
agent-browser state save ~/.slack-auth-state.json
Restore Session
agent-browser state load ~/.slack-auth-state.json
agent-browser --headed open "https://app.slack.com"
agent-browser wait --load networkidle
agent-browser get url
Actions
Read Messages
agent-browser press "Meta+k"
agent-browser wait 500
agent-browser snapshot -i
agent-browser fill @eXX "channel-name"
agent-browser wait 1000
agent-browser snapshot -i
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -c
Send a Message
agent-browser snapshot -i
agent-browser click @eXX
agent-browser type @eXX "Your message"
agent-browser press "Enter"
agent-browser wait 1000
agent-browser snapshot -c
Use type not fill — Slack's compose box is contenteditable.
Reply to Thread
agent-browser snapshot -c
agent-browser hover @eXX
agent-browser snapshot -i
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -i
agent-browser click @eXX
agent-browser type @eXX "Thread reply"
agent-browser press "Enter"
Send DM
agent-browser press "Meta+k"
agent-browser wait 500
agent-browser snapshot -i
agent-browser fill @eXX "Person Name"
agent-browser wait 1000
agent-browser snapshot -i
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -i
agent-browser click @eXX
agent-browser type @eXX "Your DM"
agent-browser press "Enter"
Search
agent-browser press "Meta+g"
agent-browser wait 500
agent-browser snapshot -i
agent-browser fill @eXX "search query"
agent-browser press "Enter"
agent-browser wait --load networkidle
agent-browser snapshot -c
React with Emoji
agent-browser snapshot -c
agent-browser hover @eXX
agent-browser snapshot -i
agent-browser click @eXX
agent-browser wait 500
agent-browser snapshot -i
agent-browser fill @eXX "thumbsup"
agent-browser wait 500
agent-browser snapshot -i
agent-browser click @eXX
Check Notifications
agent-browser snapshot -i -c
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -c
Check Unreads
agent-browser snapshot -i -c
agent-browser click @eXX
agent-browser wait --load networkidle
agent-browser snapshot -c
Critical Rules
- ALWAYS re-snapshot after every action — refs invalidate constantly
- Use
--headed mode so user can see and intervene
- Set viewport 1920x1080 — Slack hides features at small widths
- Use
Meta+k (quick switcher) as primary navigation
- Use
type not fill for compose boxes (contenteditable)
- Wait after actions:
agent-browser wait --load networkidle
- Save auth state often:
agent-browser state save ~/.slack-auth-state.json
Semantic Locator Fallbacks
When refs don't match:
agent-browser find text "general" click
agent-browser find role textbox fill "Hello"
agent-browser find placeholder "Message #general" type "Hello"
agent-browser find role button click --name "Send"
Debugging
agent-browser screenshot /tmp/slack-debug.png
agent-browser get url
agent-browser snapshot
Deep-Dive References