Claude Code Now Works in the Cloud While You Sleep

Highlights

• UltraPlan clones GitHub repositories into a cloud container to generate and execute development plans in the browser or terminal.

• A single UltraPlan planning and revision session consumed 33% of the allocated usage quota for the demo period.

• The Superpowers plugin generates plans that average 833 lines of text, significantly more detailed than the 195 lines produced by UltraPlan.

• Superpowers follows a test-driven development (TDD) approach, writing test cases for features like versioning flags before implementation.

• Creating a plan with Superpowers utilized approximately 75,100 tokens, including 1,900 tokens dedicated specifically to AI skills.

• UltraPlan requires a Pro or Max subscription and a GitHub-hosted repository to facilitate the cloud-based cloning and execution process.

Timeline

1. 00:00 - 02:11

   UltraPlan cloud environment and setup

   The system initializes a cloud session to mirror the user's project environment. This setup allows for feature planning and implementation without taxing local hardware. Users must have a specific version of Claude Code installed to access the /ultraplan command.

   • UltraPlan moves the planning process from the local CLI to a cloud-based container on the web.
   • The feature requires a GitHub repository to allow the system to clone and analyze the codebase.
   • A rainbow visual effect in the terminal distinguishes the UltraPlan command from standard operations.
2. 02:12 - 04:31

   Planning performance and usage costs

   Initial attempts may fail or misidentify repository contents, necessitating manual revisions. The interface allows users to select specific text blocks and add comments for the AI to address in secondary iterations. Implementation in the cloud environment lacks GitHub credentials by default, requiring manual branch and PR creation unless specific app permissions are granted.

   • The initial plan generation for a release pipeline took approximately two to three minutes using sub-agents.
   • Usage metrics spiked from 4% to 37% during the creation and revision of a single project plan.
   • Revised plans include visual flow diagrams and specific lists of files to modify for GitHub actions.
3. 04:32 - 06:26

   Comparison with Superpowers local plugin

   Local execution via Superpowers avoids the need for repository cloning since it has direct access to the files. It provides a more robust goal architecture and technical stack outline. Total token usage for a session reached 75.1K tokens, which may exceed standard 5-hour limit estimates depending on prompt caching efficiency.

   • Superpowers asks double the amount of clarifying questions compared to UltraPlan to ensure plan accuracy.
   • The plugin operates in two distinct phases: a design plan for requirements and an implementation plan for bite-sized tasks.
   • Test cases are written and failed intentionally before any functional code implementation occurs.
4. 06:27 - 07:40

   Ideal use cases for cloud versus local planning

   The choice between tools depends on the physical environment and the required depth of the plan. While cloud planning offers mobility, local planning facilitates a more thorough dialogue with the model. Future updates may expand these features to different subscription tiers and platforms beyond the current Pro and Max requirements.

   • UltraPlan serves developers who need to transition between machines or work from mobile devices during commutes.
   • Local tools like Superpowers are preferable for 90% of standard work due to integration with local skills and MCP tools.
   • Current access to UltraPlan is limited to the CLI, with potential for future web-native session starts.