SOP-021: Cloud Agents

Document Control

• Version: 1.0
• Last Updated: 2024
• Category: Advanced

---

Purpose

Use Cloud Agents to run coding tasks in the cloud from Slack, Linear, or the web interface.

---

Scope

This SOP covers:

• Understanding Cloud Agents
• Triggering agents from integrations
• Monitoring and managing agents
• Configuration options

---

Prerequisites

• Cursor Team or Pro subscription
• GitHub connected to Cursor
• Usage-based pricing enabled
• Integration connected (Slack, Linear, or GitHub)

---

Procedure

Step 1: Enable Cloud Agents

1. Go to Cursor Dashboard
2. Connect GitHub if not connected
3. Enable usage-based pricing
4. Confirm privacy settings

Step 2: Trigger from Integrations

From Slack:

@Cursor fix the login bug in the auth service

From Linear:

• Delegate issue to Cursor (assignee)
• Or comment @Cursor [instructions]

From GitHub:

@cursor implement this feature

From Web: Visit cursor.com/agents

Step 3: Monitor Progress

In Slack:

• ⏳ Running
• ✅ Completed
• ❌ Failed

In Dashboard:

• View all agents at Cloud Agents tab
• Check status, logs, and created PRs

Handoff to Desktop: Click "Open in Cursor" to continue locally.

Step 4: Provide Follow-up Instructions

Add more context to running agents:

In Slack:

@Cursor also add unit tests for the new function

In Linear: Comment with @Cursor [additional instructions]

Step 5: Configure Defaults

At Cloud Agents Dashboard:

Setting	Description
Default Repository	Primary repo when not specified
Default Model	AI model to use
Base Branch	Branch for PR creation
Privacy Mode	Zero data retention

---

Configuration Options

Override defaults with options:

Option	Syntax	Description
Repository	[repo=owner/repo]	Target repository
Branch	[branch=develop]	Base branch
Model	[model=claude-sonnet]	AI model
Auto PR	[autopr=false]	Disable auto PR

Example:

@Cursor [repo=myorg/myapp, branch=feature] implement dark mode

---

Agent Capabilities

Cloud Agents can:

• Clone and modify repositories
• Create branches and commits
• Open pull requests
• Run commands in sandboxed environments
• Read issue/PR context

Cloud Agents cannot:

• Access your local machine
• Modify files outside the cloned repo
• Run arbitrary network requests

---

Privacy Mode

When enabled:

• Zero data retention with AI providers
• Code not used for training
• Temporary storage while running only

Configure at Privacy Settings.

---

Best Practices

Practice	Reason
Be specific in prompts	Better results
Reference files/issues	Provides context
Use follow-ups	Refine incrementally
Review PRs carefully	Verify changes
Set repository defaults	Avoid confusion

---

Verification Checklist

• [ ] Cloud Agents enabled
• [ ] GitHub connected
• [ ] Default repository set
• [ ] Test agent completes successfully
• [ ] PR created and reviewable

---

Troubleshooting

Issue	Solution
Agent not starting	Check GitHub permissions
Wrong repository	Set defaults or use [repo=...]
Agent stuck	Check dashboard for status
No PR created	Verify branch permissions
Request ID needed	Use verbose mode for support

---

See Also

• SOP-017: GitHub Integration
• SOP-018: Slack Integration
• SOP-019: Linear Integration