Everyone who discovers Claude Code goes through the same phase. You build one skill that actually works, you feel the productivity jump, and then you try to turn every single workflow into a skill. Two weeks later you have fourteen half-broken skills, none of them get used, and you've spent more time configuring them than you ever spent doing the underlying work.
I went through this. Most operators I know went through this. The lesson at the other end is boring but real: most workflows should not be skills. The point of a skill is to compress a repeated, multi-step, file-touching workflow into a single command. If the workflow isn't repeated, isn't multi-step, or doesn't touch files, you're better off in chat.
This is the checklist I use now before I build anything new. It's saved me from at least a dozen skills I would have abandoned.
The Honest Threshold
Building a working Claude Code skill takes 30 to 90 minutes of real time. The fast end is a single-purpose wrapper. The slow end is anything that touches a third-party API, has branching logic, or needs configuration per client. Add an extra 20 to 30 minutes if you actually test it on real data instead of trusting the first run.
The break-even is somewhere around the fifth time you run the workflow. Below that, you're spending creation time you won't recoup. Above that, the skill starts paying for itself. Most workflows people try to skill-ify will be run exactly twice before getting abandoned. That's the trap.
The 4-Point Checklist
A workflow is skill-worthy if at least three of these four are YES. Two or fewer means stay in chat.
1. Is It a Repeated Workflow?
Do you run this same shape of work at least weekly? Not "I might do this again someday" — actually run it on a cadence you can name.
The Monday-morning client report I send every week is a yes. I run it 4-5 times every Monday, every week, all year. The pitch deck I built for one board meeting last March is a no. I will never run that workflow again in that exact shape, no matter how cleverly I templated it.
2. Does It Touch a File Path?
Skills shine when there's a file system component — reading from a CSV, writing a report, scanning a directory, comparing two folders. Chat is great for thinking. Skills are great for thinking plus I/O.
Auditing a Shopify product export against a Meta ad copy folder is a yes. There are file paths, structured data, and a defined output. Brainstorming a tagline is a no. There's no file involved, no path to reach for, nothing to write back. That's a chat task.
3. Does It Need Context Across Runs?
Does the workflow need persistent state — client-specific config, brand tone files, naming conventions, account IDs — that you don't want to re-type or re-paste every time?
Anything client-specific is a yes. My weekly report skill knows which clients have Amazon and which only have Shopify, which pixel ID belongs to which account, and what tone Barbara wants for the school report versus what Tom wants for the diaper brand. That state lives in the skill so I never have to re-explain it. One-off questions are a no. If you only need the context once, you don't need a skill to hold it.
4. Does It Chain Multiple Tools or Steps?
Does the workflow require three or more tools, API calls, or sequential steps? Pulling Meta data, joining it against Shopify orders, calculating churn risk, and outputting a re-engagement email list is a yes. That's four steps, three APIs, one structured output. It needs to be a skill because doing it by hand in chat means I'm copy-pasting between four windows.
"Rewrite this paragraph in a tighter tone" is a no. It's one step. It's a single chat message. Wrapping it in a skill adds friction, not leverage.
Examples — Skill-Worthy
These are the workflows on my machine where the math has obviously paid off:
- yesterday-recap — Pulls Meta spend, Google spend, Shopify revenue, and Amazon revenue for a single client and prints a cross-channel snapshot. Runs daily for 6+ clients. Touches files, holds client config, chains 4 APIs.
- monday-report — Generates the weekly dashboard email I send to clients every Monday. Pulls multi-channel data, formats an HTML email, and sends via Resend. I run it 8-10 times every Monday.
- hook-battery — Generates 10 ad hooks across 5 psychological trigger types from a brief file. I run this every time I'm about to write Meta ad copy for a new campaign. Easily 3-5x per week.
- client-onboarding — Scaffolds a new client folder, builds a research baseline, and updates the agency index. Every new client gets one. Saves 90 minutes of manual setup each time.
- The Meta Ads CLI — Not a Claude Code skill exactly, but the same idea. JSON brief in, paused campaign on Meta out. I wrote about that one separately.
Notice the pattern. Every one of these hits all four criteria — repeated, file-based, context-dependent, multi-step.
Examples — NOT Skill-Worthy (Stay in Chat)
These are the things people try to turn into skills and shouldn't:
- "Generate a tagline for X" — one prompt, no state, no chain. Just ask in chat.
- "Explain this concept to my team" — pure thinking task. No I/O. No repetition.
- "Summarize this article I just pasted" — no file path you'll reuse. No persistent context. Chat handles this in three seconds.
- One-off email drafting — if you're writing one email to one person about one situation, a skill is overhead.
- First-time research into a new topic — exploratory work, no defined output shape yet. Build a skill once you know what the output should look like, not before.
If you find yourself building a skill called something like "answer my questions about marketing," delete it. That's just chat.
The Hidden Cost of Skills
This is the part nobody mentions. Skills are an asset with carrying cost. Every skill you build is something you have to maintain.
When Meta updates their API (which they do every quarter), my reporting skills break. When Shopify changes the order export schema, my churn skill breaks. When a client moves from Klaviyo to a different ESP, three skills need updating. I budget 4-6 hours a month for skill maintenance across the stack, and that number scales with how many skills I have.
There's also drift cost. A skill you wrote six months ago for a workflow you no longer use is dead code that still appears in your skill list, takes up cognitive space, and occasionally gets triggered when you don't want it to. I prune skills every quarter the same way I prune Meta ad sets — if it didn't earn its place, it gets archived.
The 80/20 of Skill-Worthy Workflows in Marketing Ops
If you're running performance marketing and you're trying to figure out where to start, these are the categories where skills almost always pay off:
- Reporting — daily and weekly cross-channel pulls. The highest-leverage skill category by a wide margin.
- Ad creative production — hooks, primary text variants, headlines from a brief.
- Account audits — Meta account audits, Google QA checks, landing page audits.
- Client comms — weekly emails, monthly reviews, ad approval requests.
- Client onboarding — folder scaffold, account audit baseline, research kickoff.
- Performance reviews — period-over-period comparisons, post-campaign retrospectives.
- Creative briefs — converting a strategy doc into a structured brief for designers or media buyers.
Everything outside those seven categories is a coin flip. Run the checklist before you build.
The Takeaway
The right number of skills for most marketing operators is between 8 and 20. Below 8 and you're leaving leverage on the table. Above 20 and you're probably maintaining more skills than you're using.
The skills that survive my quarterly prune all share the same shape — repeated weekly, file-based, client-context-dependent, and multi-step. The ones I deleted all failed the checklist on at least two of four criteria. Run the checklist before you build, and you'll skip the fourteen-broken-skills phase entirely.
If you want the workflows that actually deserve to be skills already built, The Operator on operatorstack.app is the pre-built set covering the exact categories above (reporting, ad creative, audits, onboarding), plus The Lab where new skills land every month. One-time $397 launch price, going up as the Lab grows. Or you can hire Clare Digital and we'll build them for your accounts directly. Book a call here.