This is a good post and obviously hits a nerve with folks. I think folks are rightly concerned about 'heaviness' of skills, and I have criticised other skill libraries/frameworks for being too heavy-handed.
My goal is always to make the skills as light as they can possibly be. Not only within the skills themselves, but lowering the context load of the overall approach. Small skills are easier to maintain, audit, and survive model changes better. Recent guidance on prompting Fable seems to back that up - say less to the model, and it does better.
Documents are for the model, not the human
Here's where I think you got something wrong. Every 'document' I use is designed for the constraints of models, not for humans.
Why have a spec at all? To clearly define the destination in an artifact which can then later be checked via /code-review.
Why have tickets? So that you can split work into smart zone-shaped sizes.
I'm using the terms 'spec' and 'tickets' because they are familiar to devs - not because I am trying to mimic Jira.
Wayfinder, similarly, is designed for the constraints of models. It breaks down planning into context-window shaped chunks of work.
The smart zone
If there is a bitter lesson in AI coding, it's about the "smart zone". My skills assume that working with irrelevant information in your context window is bad, and so they break work into session-sized chunks via /to-tickets.
This is strictly an AFK agent optimization - AFK agents do better with chunks scoped ahead of time. I've experimented with just a spec and /goal, but the results were worse.
As you can tell, your itch is something I think about every release of my skills. I am always trying to reduce my maintenance surface, to make the skills smaller and more useful. I often do a no-op pass on my skills to check for things the models now do out of the box.
Thanks for giving me an opportunity to articulate this! A really well-phrased question.
Bruh... someone recreated the UI of Claude Code, Codex & Grok as reusable ShadCN components. ��
This open-source library lets you drop beautifully recreated AI interface components straight into your own apps or CLI projects.
You'll find components for:
→ prompts & chat messages
→ tool calls
→ code diffs
→ permission dialogs
→ todo lists
→ slash commands
→ and much more...
Perfect if you're building AI apps, terminal UIs (TUIs), or CLI-based developer tools and don't want to build everything from scratch :)
Source 🔗: github(DOT)com/theswerd/brainless
Follow me for more amazing AI, Coding & Web Dev insights 💎
Here's a step-by-step process to kill all the bloat from your Claude Code system prompt:
1. Run a proxy so you can see exactly what gets sent to Claude Code (included in the article)
2. "Fuck, there is so much cruft in there"
3. Use my settings.json to kill all the bloat
Down to a clean 13K tokens to start each session with. Nice.
Full process here:
https://t.co/Kie84HMe2A
mattpocock/skills v1.1 is out!
- /wayfinder helps you plan more ambitious work than ever
- /to-spec and /to-tickets replace /to-prd and /to-issues
- /implement + /code-review complete the whole lifecycle
- /research and /prototype help support wayfinder, or can be used independently
- Crucial fixes to /grill-me
Run npx skills@latest add mattpocock/skills to update!