A team's /analyze-deps skill reads dozens of files and prints a detailed report on each one. Developers run it several times during long sessions, and afterwards the main conversation loses track of earlier decisions under all the output. What is the most effective fix?

Answer: C — Add context: fork to the skill's frontmatter so it runs in an isolated sub-agent context

By default a skill runs inline, so everything it reads and prints stays in the main conversation. context: fork runs it in an isolated sub-agent context and returns only its result, which suits verbose skills like codebase analysis. (A) limits a useful tool without fixing its output. (B) loads the instructions in every session and doesn't isolate anything. (D) restricts which tools it can use, but its output still lands in the main conversation.