Every sprint you spin up the same kind of worker with the same instructions, so you write it up as a custom subagent and commit the file. A colleague then asks whether moving that file somewhere else would let it use more tools. What should you tell them?

The principle — two questions about a custom worker look like one and are answered in two different places: who may reach it, and what it may reach for.

Why the key is correct — Anthropic separates them in a single sentence: a subagent's file location determines who it's available to, and its frontmatter determines what it can do. So moving the file changes the audience and nothing else. The reason the file exists at all is the sprint pattern in the stem — define a custom subagent when you keep spawning the same kind of worker with the same instructions.

Why the others are wrong — location cannot widen capability, because capability is declared rather than inherited from a path. Nor is tool access fixed: it is one of the things the definition sets, which is why moving the file is the wrong lever rather than the right one. And Claude choosing which worker runs is a separate decision from what that worker may then do.

Remember this — path decides who, definition decides what. Moving a file answers the wrong question.

Sources — Anthropic's Claude Code subagents documentation.