Agent Sandbox Branch-Workflow
Overview
A throwaway build exercising an agent-cloud sandbox platform’s branch-and-work model. The repo is a minimal scratch workspace (chucky.json manifest plus a set of generated text artifacts) used to validate how an AI coding agent creates branches, performs work, and emits reports.
Why It Exists
To kick the tires on a Claude Code-style cloud sandbox for developers, confirming the developer experience of agent-driven branches, task execution, and automated artifact generation before building anything on top of it.
What We Built
A bare project defined by a chucky.json manifest (name: build-branches, “A build for branches and work”) and several agent-produced outputs, including a research report.txt summarizing the sandbox platform’s capabilities (cloud sandbox for Claude AI integration, PHP SDK, etc.) and assorted generated text files (poem, notes). Scope is intentionally tiny: an evaluation harness, not an application.
Technologies & Approach
Built around an agent-cloud sandbox platform configured via a single JSON manifest, with the agent operating on git branches and writing results to the workspace. Deliberately minimal to isolate the platform’s branch/work mechanics.
Outcome / Impact
Validated the agent sandbox’s branch-based workflow and automated report generation, informing the studio’s broader agent-orchestration direction. Honestly framed as a one-day evaluation spike.
Capabilities Demonstrated
- Hands-on evaluation of agent-cloud sandbox platforms
- Branch/workspace-driven agent workflow validation
- Automated research and artifact generation via agents