Core concept: Skill behaviour comes from real frontmatter fields, and only real ones. Two needs are stated here: an isolated context, and a read-only tool set. Two documented fields cover them.

Why B and C are correct:

B: The skills docs say it directly: "Add context: fork to your frontmatter when you want a skill to run in isolation." The forked run "won't have access to your conversation history".
B: Only the outcome comes back, because the subagent "summarizes its results and returns them to your main conversation". The exploration noise never enters the main window.
C: allowed-tools is the field that names a skill's tools: "Tools Claude can use without asking permission during the turn that invokes this skill." Listing Read, Grep, Glob scopes the skill to reading and searching, with no Write, Edit or Bash.
C: Pair it with agent: Explore for a read-only agent, and use disallowed-tools when a write tool must be removed outright.

Why the others are wrong:

A: There is no isolation field. The frontmatter reference lists name, description, allowed-tools, disallowed-tools, model, context, agent and more, but never isolation.
D: There is no permissions field either. Permission rules live in your settings file as allow, ask and deny rules, not in SKILL.md.

Remember: context: fork for isolation, allowed-tools for the tool list. Invented keys such as isolation and permissions do nothing.

Question 191