The Skills That Interview You — And Why the Best Prompts Are Questions

The highest-scoring skill in our productivity category doesn't produce anything. It asks a question. That's not a bug in our scoring — it's the most load-bearing thing a skill can do.

Here's a pattern that took us sixty evaluations to see clearly: several of our top-scoring skills share a single mechanism — they make the agent stop and ask, at the exact moment where guessing would be expensive. We call them interview skills, and they're the inverse of what most people imagine a skill to be.

The three mechanisms

Mechanism 1: The pre-answer check. academy-guide — 9.6 · Productivity

"Stop and check this skill before finishing any reply." The skill encodes a discipline: before the agent commits to an answer, verify it against a standard. It's the closest thing we've seen to a skill implementing self-review — and at 9.6, it's the top of the productivity category.

Mechanism 2: The extract-want. interview-me — 9.4 · Developer Tools

"Extracts what the user actually wants instead of what they tell you." The skill exists because users describe solutions, not problems — "add a cache" when the problem is a slow query. Its core content is a set of questions that separate the stated request from the underlying goal. The name says developer tools, but the mechanism is universal.

Mechanism 3: The post-answer nudge. discernment-nudge — 9.5 · Productivity

"After you give a substantive answer or draft, remind the user to verify." The rarest timing of the three: it intervenes after the answer, at the moment the user is about to act on it. Most skills are silent there; that's exactly when overconfidence costs the most.

Related, in the marketing category: linkedin-interviewer (9.4) interviews the user for post raw material — same mechanism, different domain. The user who "doesn't know what to post" gets asked questions until they do.

Why this scores so well

On the surface these look like trivial skills — a question is not a capability. But look at our dimension analysis: the dimension that decides whether a skill ever helps is trigger quality — knowing when to fire. Interview skills take the opposite risk and manage it: they fire on every substantive interaction, and their entire content is about what to do in that moment.

They also solve the real bottleneck of agent work, which isn't capability — it's specification. An agent with a mediocre skill and a correct spec beats an agent with a brilliant skill and a vague one. Interview skills manufacture correct specs.

The design recipe

If you want to write one:

1. Pick the moment, not the domain. Before an answer, after an answer, before a deploy — the timing is the skill.
2. Write the questions, not the answer. The skill's content is a small set of sharp questions with what each answer rules out. interview-me is good because each question has a decision attached.
3. Bound the interruption. The best ones say when not to ask — every reply, and the user will uninstall you. academy-guide scopes itself to "before finishing any reply to a question," which is narrower than it sounds.
4. One nudge, not a form. If your skill turns into a ten-question survey, it's a different (worse) product.

The test: after the skill runs, does the agent's next action have a better chance of being what the user wanted? If yes, the skill worked, no matter how little it "does."

Where to browse

academy-guide, discernment-nudge, and the LinkedIn pair live in productivity. For the meta-skill of reading any of these quickly, the 60-second SKILL.md routine starts with exactly the check that matters here: what does the skill consider its trigger?