Skip to main content
Connect Slack to run agent scenarios without leaving your workspace. Mention @hud with a command and get results posted back to the channel.

​
Setup

  1. Go to hud.ai/project/settings
  2. Find the Integrations section
  3. Click Add to Slack
  4. Authorize HUD for your workspace
Once connected, you’ll see the workspace name and a “Connected” badge.
You need admin or owner permissions on your HUD team to set up integrations.

​
Commands

​
Run a Scenario

@hud run <environment> <scenario> [arg=value ...]
Runs an agent on a specific scenario. Arguments with spaces must be quoted. Examples: 
@hud run browser-env search-task query="best restaurants in SF"

@hud run code-sandbox debug-task file=main.py line=42

@hud run ops-diagnostics investigate issue_url="https://sentry.io/issues/12345"
The bot reacts with ⏳ while processing, then ✅ on success with a trace link to follow the run.

​
Example: Ops Debugging Bot

We use this internally at HUD—when a Sentry alert fires, someone pastes the issue URL into our #ops channel: 
@hud run ops-diagnostics investigate issue_url="https://sentry.io/issues/12345"
The agent:
  1. Connects to Sentry via MCP
  2. Pulls error details, stack traces, and context
  3. Analyzes the root cause
  4. Posts a summary with suggested fixes
The whole team sees the diagnosis without anyone needing to context-switch into Sentry. We’ve found it cuts our initial triage time significantly.
See the Ops Diagnostics Cookbook to set up a similar environment for your team.

​
List Available Scenarios

@hud list
Shows all environments and scenarios you have access to.

​
Get Help

@hud help
Shows available commands and usage examples.

​
Options

You can customize runs with additional parameters:
ParameterDescriptionDefault
modelModel to use (after scenario name)claude-sonnet-4-5
max_stepsMaximum agent actions100
Example with options: 
@hud run my-env my-scenario claude-opus-4-5 50 query="hello world"
This runs my-scenario with claude-opus-4-5, max 50 steps, and query="hello world".

​
Viewing Results

When you run a scenario, HUD immediately responds with:
  • ✅ Started confirmation (or ❌ if there’s an error)
  • Trace link — Click to follow the agent’s progress live
The trace page shows detailed tool calls, observations, and updates in real-time as the agent runs. Once complete, you’ll see the final evaluation result and reward.

​
Finding Commands from the Platform

Every scenario card on the platform has a code snippet panel. Click the </> icon, then select the Slack tab to see the exact command for that scenario with proper argument formatting.
Copy the command directly from the platform for any scenario.

Slack command in scenario card

​
Troubleshooting

​
”I don’t see @hud in my workspace”

Make sure you’ve completed the OAuth flow in Settings. You may need to refresh Slack after authorization.

​
”Permission denied” errors

The person running the command needs access to the environment in HUD. Check your team permissions on the platform.

​
”Scenario not found”

Double-check the environment and scenario names. Use @hud list to see available options.

Environments

Create environments with scenarios

Models

Available models for agent runs