context: fork

context: fork is a SKILL.md frontmatter value that runs the skill in isolation, in a forked subagent context. The skill's CONTENT becomes the prompt that drives the subagent, and it does not have access to your conversation history. The companion field agent picks which subagent type executes it.

Use context: fork when the skill's work would flood the main conversation, or when you want the skill to start from a clean slate rather than inherit whatever is already in context. A documented consequence worth knowing: the built-in Explore and Plan agents skip CLAUDE.md and git status to keep their context small, so a forked skill using agent: Explore sees only the SKILL.md content and the agent's own system prompt — not your project instructions. That is a feature if you want isolation and a trap if the skill quietly depended on CLAUDE.md.

Interaction with slash-command stacking: expansion stops at the first token that is not an inline user-invocable skill, so a skill that runs as a forked subagent ends the run there. The inverse pattern — a custom subagent that uses skills as reference material — is the subagent skills preload field instead. This isolation is a key aspect of how a subagent operates, distinct from how skills might be used with progressive disclosure or within a larger context-window.