Claude, Listener Creator.

It's a reasonable assumption to make: if you ask Claude to flag and forward certain emails, something must be running in the background, constantly scanning your inbox. It isn't. The listener-creator skill reveals something more specific, and more limited, about how a Claude Skill actually works.

A folder, not a watcher

Open the listener-creator skill and there's no watcher process running anywhere. There are just two things: a SKILL.md file, about nine kilobytes of plain text, and a templates folder sitting beside it. Claude reads SKILL.md before it acts, then writes back a listener definition based on what it finds there. The instruction set is the file itself, not a background process quietly running on your account.

Steps, read top to bottom

Inside SKILL.md is a Steps section that Claude works through in order: read the request, write the condition, return the definition. Ask for a listener that flags any email from your boss marked urgent and forwards it immediately, and listener-creator writes exactly that condition into a file. From that point on, whenever a new email matches both boss and urgent, the same forward fires, every time, because the condition written into the file doesn't change on its own.

Reliable, and precisely bounded

That fixed condition is exactly what makes a listener reliable, and exactly what limits it. Because the rule never drifts, every email from your boss marked urgent forwards the same way, consistently. But an equally urgent email from a client never triggers it, because "client" was simply never written into the condition in the first place. A listener matches the literal words stored in its file, not a judgment call about how urgent a given email actually feels. It's a file naming one condition, firing on exactly that condition every time, and staying silent on anything that only resembles it.

Key takeaways

• A Claude Skill listener is a file describing one condition, not a live process monitoring your inbox.
• The listener-creator skill itself is just two files: a SKILL.md instruction set and a templates folder.
• Claude works through a fixed Steps section: read the request, write the condition, return the definition.
• A listener's reliability comes from its condition never changing on its own, which is also its limitation, since it only matches what was actually written into the file.

Try it yourself

Ask Claude: "I want a listener that flags any email mentioning a contract renewal and forwards it to my legal team. Read the listener-creator skill and walk me through what you'll do before you do it, show me exactly which condition you're writing into the file." Watching Claude name the condition before writing it is a direct way to see the Steps section working on your own request, narrated by Liam, in for Bear, as part of Humanitarians AI's Claude Basics series.