From Session to Skill
I asked Claude to fix all typographic errors in a Dutch InDesign document. Twenty minutes later, we had a clean document—and a reusable skill that allowed you to repeat most of it on the next document. The journey between those two points reveals both the promise and the current limitations of turning AI sessions into lasting automation.
The Session
The document was a Dutch text riddled with typical typographic issues: wrong dashes, incorrect apostrophes, missing thin spaces, disabled hyphenation. Using Sidekick for InDesign, I sent Claude to work.
Claude’s approach was methodical. It researched Dutch typographic conventions online, analyzed the document character-by-character to find hidden control characters, then began making corrections systematically—converting hyphens to en-dashes, fixing spacing, applying subscripts to chemical formulas, setting acronyms in small caps.
Impressive. But a single prompt wasn’t enough.
It’s tempting to assume that if Claude responds confidently, it understood the full scope of what you asked. It didn’t. Dutch typography has edge cases that don’t appear in general style guides. Claude doesn’t know what it doesn’t know—and neither do you, until you check.
The Handholding
When I examined the results, several things were still wrong. The apostrophe in “‘s avonds” curved the wrong direction. Paragraph indents had appeared unexpectedly. The percentages lacked thin non-breaking spaces. Hyphenation was still set to English.
Each issue required a nudge: “The apostrophe is wrong—check what it should be.” “The indent you added is wrong.” “Use non-breaking thin spaces.” “Make sure hyphenation is set to Dutch.”
Claude fixed each problem when pointed out. But it couldn’t see what I could see—the trained eye catching a 6-shaped quote where a 9-shaped apostrophe belonged, or noticing that line breaks fell in un-Dutch places.
This is the reality of AI-assisted design today: genuinely useful, but not autonomous. You’re collaborating, not delegating.
The Skill
After the session, we had accumulated real knowledge: the execution order, the GREP patterns, the edge cases. I asked Claude to generalize the rules and package them as a skill—a structured knowledge file that Sidekick can use on future documents.
The result: dutch-typography.skill, containing comprehensive Dutch typographic rules, ready-to-use patterns, and an automation script.
The Gap
Here’s the interesting part: when we later used the skill on a new document, it missed some of what we’d discovered in the original session. Rules that Claude had applied through our back-and-forth conversation didn’t all make it into the packaged skill. The apostrophe fix that required multiple attempts? The skill included the pattern, but the same edge case might trip it up again.
This is an area that needs more research. How do we ensure that all the corrections from an interactive session—including the ones that required human prompting—get encoded into the skill? How do we capture not just the rules but the verification steps, the fallback approaches, the “if this doesn’t work, try that” knowledge?
Skills are promising. They compound learning. But right now, there’s signal loss between a successful interactive session and the skill that emerges from it.
What This Means
The real value isn’t just the clean document—it’s the skill that comes out the other end.
Sidekick for InDesign makes it possible to tackle specialized tasks like Dutch typography correction conversationally. You don’t need to know InDesign scripting—you describe what you want, and Claude figures out how to do it. But expect to collaborate, not just command. Watch the output. Point out what’s wrong. Guide the process.
Then package what you learned as a skill. The first document takes twenty minutes of back-and-forth. The next one? Seconds. That’s the promise: every session can become reusable knowledge.