Best practices, workarounds, and feedback while working with the Cloud Agent

Because the Cloud Agent is currently in beta, some functionality is still being developed. Below are some layout tips and user workarounds to optimize your generated diagrams:

Scale & scope
The agent performs best when given a clear scope. Helping the agent focus on specific targets ensures faster, highly accurate data retrieval within optimal context windows.

Tip: Prompt with high specificity. Use exact tags, accounts, and regions rather than relying on generic prompts.

Optimizing diagram layouts
Like any complex model, the agent is thorough by design — it maps every resource and relationship you ask for. Rendering a broad architecture all at once can produce a dense layout that's hard to read at a glance.

Tip: Keep your diagrams clean and readable from the start:

• Narrow the scope: Focus your initial prompt on a single region, VPC, or key application rather than your full environment.
• Use containers: Ask the agent to group resources into collapsible containers so you can expand only what you need.
• Simplify connections: Ask the agent to omit non-critical connection lines for shared resources to maintain visual flow.

Editing diagrams
The agent builds each diagram with full awareness of your cloud data in that session. Starting a new prompt to edit an existing diagram opens a fresh session without that context, which can produce unexpected results.

Tip: Instead of asking the agent to modify an existing diagram, describe what you want in a new prompt. Treat each diagram as a starting point — refine by regenerating with a more targeted request rather than editing in place.

Facts vs. inferred relationships
The agent maps both confirmed resource data and logical relationships between services — but not every connection in your architecture is explicitly declared in your cloud config. Some relationships are inferred based on common patterns.

Tip: Add "distinguish between confirmed and inferred relationships" to your prompt. The agent will call out which connections are based on imported data vs. reasonable assumptions, so you can validate before sharing the diagram with your team.

Feedback

We are seeking feedback while we continue to improve the Cloud Agent. Comment below with any feedback, specifically:

• Data query context exhaustion: Reports on the frequency of the agent exceeding its context window when scanning expansive cloud environments.
• Rendering and tool errors: Instances where the diagram generation step fails to execute despite a successful initial data fetch.
• Edit-in-place regressions: Notable examples where manual edits caused a loss of essential metadata or layout integrity, particularly while permanent regeneration fixes are pending.
• Resource hallucinations: Feedback regarding scenarios where the agent visualized non-existent assets or services not found in your actual cloud dataset.
• Prompt reliability trends: Successes or failures with specific phrasing; we are cataloging these to improve our broader prompting recommendations.