context-fork-skill

You can isolate testing tasks by creating a forked sub-agent context with its own independent conversation history. This prevents cross-contamination, ensuring the main context remains clean and unaffected by the isolated execution environment.

A forked sub-agent context is used to execute complex, multi-step workflows in an isolated environment. It maintains separate state and conversation history, preventing task-specific data from leaking into the primary agent context.

Yes, you can enforce controlled tool access by configuring allowed_tools for the forked context. This ensures separate state management and controlled resource usage when executing specialized sub-agent tasks like code analysis.

Yes, isolated sub-agent execution supports configurable agent types such as general-purpose, Plan, Explore, and code-reviewer. This allows you to tailor the forked context for specific complex analyses or testing workflows.

The best way to run safe experiments is by initiating a forked context for isolated task execution. This provides a separate environment with its own history, allowing safe testing without cross-contamination into the main conversation.

You should not use a forked context for simple tasks that do not require separate state or isolated conversation history. Forking is specifically designed for complex, multi-step analyses and specialized workflows where avoiding leakage is critical.