Studio 01 · The Builder Bench

Write your first research skill.

Seven small fields. Most take 1-3 sentences each. By the end of this studio you'll have a real research skill you can hand to Claude — or paste into a future conversation as context. Your answers save automatically in your browser. Export the whole thing as Markdown when you're done.

👩‍🔬

Marie thinks…

This is where we stop talking and start building. Pick a real recurring task from your research week. The XRD pattern you identify every Friday. The reviewer comment you've been avoiding. The control experiment you keep designing from scratch. Fill the form below for that one. Don't worry about getting it perfect — the first draft is supposed to be rough. Run it on Monday, see what fell out, refactor.

Quick start

Load a worked example, or start from scratch.

Three pre-filled research skills to look at the shape of a finished one. Pick one to load it into the form, then edit toward your real task. Or skip and start blank.

Field 01

Skill name

Kebab-case. Short, verb- or noun-led handle. Avoid clever puns.

Why it matters → The name is the handle you'll use when chaining or referencing the skill. literature-cartographer beats PaperBot9000.

Examplereviewer-comment-classifier

If you can't name it in three words, you don't know what it is yet.

Field 02

Input (one line, structured)

What does the skill consume? Schema-ish, not prose. List the fields it expects.

Why it matters → If you can't write the input as fields, the skill isn't ready to run. Writing it forces you to be honest about what's actually available.

Example{manuscript_section: string, reviewer_comment: string}

Field 03

Output (one line, structured)

What does the skill produce? Make it consumable by a downstream skill — types, not vibes.

Why it matters → The output shape determines what stacks on top of this skill. Vibe outputs orphan the skill; structured outputs let it chain.

Example{classification: 'substantive'|'cosmetic'|'wrong', suggested_action: string, draft_response: string}

Field 04

Purpose (one sentence)

In one sentence: what does this skill do, and what does it produce?

Why it matters → If you need two sentences with an "and," you have two skills pretending to be one. Cut.

ExampleGiven a reviewer comment on a manuscript section, classify the comment as substantive / cosmetic / wrong, and draft a response that addresses it without sounding defensive.

Read your sentence aloud. If a colleague would say "wait, what?" the sentence isn't right yet.

Field 05

Logic — numbered moves (one per line)

3 to 6 numbered steps. Each verb-led. Each one you could ask "did I do this?"

Why it matters → This is where you transfer your judgment into the skill. Don't write "be thorough." Write what a thorough version actually does, in order.

ExampleRead the reviewer comment in context of the section it refers to. Classify: substantive, cosmetic, or wrong? If substantive: name the experiment or concession needed. If cosmetic: quote the new wording. If wrong: clarify with evidence already in the manuscript. Thank the reviewer in the opening line regardless.

If a step starts with "be" or "consider" or "think about" — rewrite it as a verb.

Field 06

Output schema (YAML, free-form)

Spell out the output shape with types. The schema is what makes the skill auditable run-by-run.

Why it matters → Without a schema, every run produces a slightly different shape. With one, you can chain reliably and a downstream skill knows exactly what to consume.

Exampleresponse: classification: substantive | cosmetic | wrong suggested_action: string draft_response: string # 2-4 sentences manuscript_location: string

Field 07

Why this ingredient matters

One short paragraph. What does this skill solve that gut-feel doesn't? What downstream skill consumes it?

Why it matters → If you can't answer this in a paragraph, the skill is probably an orphan. Or it should have been a function instead.

ExampleThe review cycle is where many manuscripts die, not from bad science but from defensive writing. Forcing the classification up front prevents arguing with a comment that should have been conceded.

Your skill, ready to ship.

Autosaved as you type. Export as Markdown to drop into a Claude conversation, or download as a .md file to keep alongside your other skills.

Your answers will save here automatically.

You finished the workshop

👩‍🔬

First Skill Built

You decomposed the task. You wrote the schema. You compressed the moves. Now run it on a real example, see what falls out, refactor. That's the plasticity loop.