Thanks for this! After reading this, I realized what I had created as a project (weekly meal plan generator) might do better as a skill within the project. But I wasn't sure: the boundary between project and skill is still a little squishy for me. So I started instead by asking Claude whether/how I should adapt it to a skill. I had a great experience with its response, pasting in case this thought process is useful for anyone else:
The Clear Split
What should stay in the Project:
Banned foods list
Available equipment
Food preferences by category
Past meal examples you've liked
Flavor brainstorms and meal ideas
Any evolving preferences or discoveries
What should move to a Skill:
The thorough instructions for generating the plan
The specific output format (plan structure, shopping list format, batch cooking steps, recipe ordering)
The workflow: analyze constraints → generate balanced meals → consolidate shopping list → create batch cooking schedule → format recipes chronologically
Consistent formatting rules
The Benefits
With this separation, you'd get:
Simplicity: Just say "generate new weekly plan" and the Skill handles the entire workflow
Consistency: Every plan follows the same structure without you re-explaining it
Cleaner Project: Your Project becomes a growing knowledge base of preferences and examples, not cluttered with procedural instructions
Reusability: If you ever wanted a separate meal prep project (say, for travel meals or meal prep for someone else), you could reuse the same Skill with different context
Maintenance: When you want to adjust the format or workflow, you update the Skill once rather than repeating instructions weekly
GREAT explanation! One of your greatest skills is how you break things down.
I found your newsletter the day after Anthropic announced Skills, and as I read through your archive I kept thinking about all the things you mentioned which were the precursors to skills. Like you said, it's nothing that a lot of us weren't already doing, but I'm glad there is an official home for them and that they will hopefully run more efficiently.
The SEO Content Optimizer example - reducing a 2-hour process to minutes by packaging instructions into Skills - shows the compounding value of invested automation. Most people build workflows and move on. The ones who package and reuse them get exponential return. Skills as portable Markdown files is elegant because it works across ecosystems. That's the opposite of lock-in.
Can you setup skills in Claude and then pull them in Claude code?
I have different spaces that I need to manage (I'm building a whole ecosystem). And I'm doing it in Claude Code. Do you know if I can reuse skills cross products?
I've experimented with Claude skills the past days and I totally see the potential for my work.
I created some very sophisticated skills but the results are very inconsistent at the moment - especially the visual parts.
One skill should output a nicely formatted and branded PDF file and while the content is generated very consistently, the PDF formatting is more like... art? 😅
So even giving it clear advice on the formatting it still looks different every time.
Maybe I should say goodbye to how I imagine my branding to look like and start putting out stuff that is truly unique! 😄
yes i'm doing this too for my newsletter dashboard visualization and strategic analysis. I created a frontend framework to follow, so whenevr I upload excel sheet consists of newsletter data, it will generate a visual dashboard with strategic analysis and improvement recommendation.
so far it's quite accurate to follow the instruction.
I guess u need to be very precise about the output, including coloring, framework, template, etc.
Now, i'm getting addicted to create many Skills and put myself out of work entirely lol
I have now created a second branding skill and it works but now I have the problem that I hit the context window when I let all two skills run in one chat.
It's hard to debug because I don't see how many token are actually used by one skill.
I guess my content generation skill that uses my project knowledge to create a custom Gene Keys profile for a client uses so many tokens, that the second branding skill for the PDF is simply too much.
It really would be time for Anthropic to get the context window up from 200k to at least 500k or better 1M. Otherwise skills like mine quickly reach their limit.
You nailed the architecture of how Claude got here. The breakdown of eras is sharp.
Here’s the tension though. Skills automate the “how,” but they still take the “why” on faith. They execute beautifully without understanding when the job itself no longer matters. Until we solve for that, we’re just scaling efficiency, not intelligence.
Still, credit where it’s due. This feels like a genuine step forward in the evolution from prompt engineering to process design.
it's all comes down to us and how this can help us at broader picture. And yes this is just for efficiency for doing repeated tasks, nothing actually fancy as AI doesn't make any decisions here, we are.
Couldn't agree more with what you said below. This is the key to using AI in general. Great post, Wyndo!
"They’re only as good as your understanding of your own workflows. If you can’t clearly articulate the steps, edge cases, and decision points in your process, you can’t build a good Skill."
Claude Skills are a game-changer. Turning repeated, specialized workflows into reusable execution modules is exactly what makes AI truly productive; less context-switching, more consistent results, and real time saved. This convergence of Projects, MCP, and Code finally makes AI practical for end-to-end execution.
this is a brilliant analysis of claude's evolution...shifting from a prompting playbook to reusable, specialized systems significantly elevates AI's utility.
Thanks for this! After reading this, I realized what I had created as a project (weekly meal plan generator) might do better as a skill within the project. But I wasn't sure: the boundary between project and skill is still a little squishy for me. So I started instead by asking Claude whether/how I should adapt it to a skill. I had a great experience with its response, pasting in case this thought process is useful for anyone else:
The Clear Split
What should stay in the Project:
Banned foods list
Available equipment
Food preferences by category
Past meal examples you've liked
Flavor brainstorms and meal ideas
Any evolving preferences or discoveries
What should move to a Skill:
The thorough instructions for generating the plan
The specific output format (plan structure, shopping list format, batch cooking steps, recipe ordering)
The workflow: analyze constraints → generate balanced meals → consolidate shopping list → create batch cooking schedule → format recipes chronologically
Consistent formatting rules
The Benefits
With this separation, you'd get:
Simplicity: Just say "generate new weekly plan" and the Skill handles the entire workflow
Consistency: Every plan follows the same structure without you re-explaining it
Cleaner Project: Your Project becomes a growing knowledge base of preferences and examples, not cluttered with procedural instructions
Reusability: If you ever wanted a separate meal prep project (say, for travel meals or meal prep for someone else), you could reuse the same Skill with different context
Maintenance: When you want to adjust the format or workflow, you update the Skill once rather than repeating instructions weekly
Love this! you really nailed it.
Aligned with benefits list you covered here too.
A lot of possibilities can be unlocked from here, happy exploring then!
Automation works best when it forces clarity before efficiency.
really good framing yeah, clarity is much important than efficiency
GREAT explanation! One of your greatest skills is how you break things down.
I found your newsletter the day after Anthropic announced Skills, and as I read through your archive I kept thinking about all the things you mentioned which were the precursors to skills. Like you said, it's nothing that a lot of us weren't already doing, but I'm glad there is an official home for them and that they will hopefully run more efficiently.
Much appreciated, thanks, Amanda. :)
I’m glad I kept up with Anthropic’s updates over time; eventually it made sense to write this in a more in‑depth way.
It feels like everything led to this moment, and we’re here to leverage every ounce of it.
The SEO Content Optimizer example - reducing a 2-hour process to minutes by packaging instructions into Skills - shows the compounding value of invested automation. Most people build workflows and move on. The ones who package and reuse them get exponential return. Skills as portable Markdown files is elegant because it works across ecosystems. That's the opposite of lock-in.
This is neat!
Can you setup skills in Claude and then pull them in Claude code?
I have different spaces that I need to manage (I'm building a whole ecosystem). And I'm doing it in Claude Code. Do you know if I can reuse skills cross products?
Thank you!
thats exactly how I edit my skills.
build them in Claude web, then open the file inside Claude Code. Edit and improve from there.
I've experimented with Claude skills the past days and I totally see the potential for my work.
I created some very sophisticated skills but the results are very inconsistent at the moment - especially the visual parts.
One skill should output a nicely formatted and branded PDF file and while the content is generated very consistently, the PDF formatting is more like... art? 😅
So even giving it clear advice on the formatting it still looks different every time.
Maybe I should say goodbye to how I imagine my branding to look like and start putting out stuff that is truly unique! 😄
yes i'm doing this too for my newsletter dashboard visualization and strategic analysis. I created a frontend framework to follow, so whenevr I upload excel sheet consists of newsletter data, it will generate a visual dashboard with strategic analysis and improvement recommendation.
so far it's quite accurate to follow the instruction.
I guess u need to be very precise about the output, including coloring, framework, template, etc.
Now, i'm getting addicted to create many Skills and put myself out of work entirely lol
I have now created a second branding skill and it works but now I have the problem that I hit the context window when I let all two skills run in one chat.
It's hard to debug because I don't see how many token are actually used by one skill.
I guess my content generation skill that uses my project knowledge to create a custom Gene Keys profile for a client uses so many tokens, that the second branding skill for the PDF is simply too much.
It really would be time for Anthropic to get the context window up from 200k to at least 500k or better 1M. Otherwise skills like mine quickly reach their limit.
You nailed the architecture of how Claude got here. The breakdown of eras is sharp.
Here’s the tension though. Skills automate the “how,” but they still take the “why” on faith. They execute beautifully without understanding when the job itself no longer matters. Until we solve for that, we’re just scaling efficiency, not intelligence.
Still, credit where it’s due. This feels like a genuine step forward in the evolution from prompt engineering to process design.
true!
it's all comes down to us and how this can help us at broader picture. And yes this is just for efficiency for doing repeated tasks, nothing actually fancy as AI doesn't make any decisions here, we are.
Couldn't agree more with what you said below. This is the key to using AI in general. Great post, Wyndo!
"They’re only as good as your understanding of your own workflows. If you can’t clearly articulate the steps, edge cases, and decision points in your process, you can’t build a good Skill."
thanks Tam! appreciated!
Claude Skills are a game-changer. Turning repeated, specialized workflows into reusable execution modules is exactly what makes AI truly productive; less context-switching, more consistent results, and real time saved. This convergence of Projects, MCP, and Code finally makes AI practical for end-to-end execution.
this is where Claude shines at the moment!
Thank you for writing this! I learned so much.
this is a brilliant analysis of claude's evolution...shifting from a prompting playbook to reusable, specialized systems significantly elevates AI's utility.
Thanks for sharing this deep guide.
wild how much Claude has evolved so far!
Great writeup. To clarify is skills only available on paid accounts with Claude? thanks
thanks! yes it's only avail for paid account at the moment
oh ok Thanks for clarifying
That's a brilliant breakdown Wyndo, as always 🎉💪🤗
thanks Karo! and also for putting comprehensive curated list in your post! :)
Great post and clear explanations. Keep up the good work!
appreciated, thank you!
I need to try this..!!
curious to see what you'll end up building Skill for
This is very helpful. Thank you.
Appreciated, thank you!