Cursor Cloud Agent Telemetry in Beacon

What is Cursor Cloud Agents?

Cursor Cloud support records cloud-agent activity after project hooks become active and uploads per-run snapshots to customer-managed storage.

Collection path: Cloud sandbox hooks with direct GCS or S3 upload

Authoritative date: June 10, 2026 from v0.0.52 in the Beacon CLI changelog.

How Beacon reads Cursor Cloud Agents

Beacon prints setup and hook content for Cursor Cloud. Project hooks are committed before the agent starts, then sandbox-local hooks capture supported prompts, tools, shell commands, files, subagents, and compaction events.

The result is normalized into Beacon's endpoint event schema so security teams can investigate activity by harness, repository, session, user, model, command, file, tool, MCP server, or approval context where the runtime exposes those fields.

Telemetry coverage

• Follow-up prompts after project hooks become active
• Tool, shell command, file, subagent, and compaction telemetry
• GCS or S3 upload paths
• Per-run object-storage snapshots keyed by Cursor conversation id

What you can do now

• Capture Cursor Cloud work that happens outside the developer endpoint.
• Keep cloud-agent telemetry in customer-managed GCS or S3.
• Use committed project hooks for repeatable cloud collection.

Get started

1. Run beacon cloud cursor print-setup.
2. Run beacon cloud cursor print-hooks and commit the generated hooks.
3. Validate that the cloud run uploads its runtime JSONL snapshot.

Source references

• Runtime docs
• Source reference