latand6

I'm a PhD student in structural engineering. My dissertation topic is about using LLM agents in automating FEA calculations on common Ukrainian software that companies use. I'm writing my literature review now and I've vibecoded a personal local dashboard that helps me manage the literature review process.

I use LLM agents to fill up the LaTeX template (to automate formatting, also you can use IDE to view diffs) in github repo. Then I run ChatGPT Pro to collect all relevant papers (and how) to my topic. Then I collect the ones available online, where the PDFs are available. I have a special structure of folders with plain files like markdown and JSON.

The idea of the dashboard is the following: I run the Codex through a web chat to identify the relevant quotes — relevant for my dissertation topic — and how they are relevant, it combines them into a number of claims connected with each quote with a link. And then I review each quote and each claim manually and tick the boxes. There is also a button that runs the verification script, that validates the exact quote IS really in the PDF. This way I can collect real evidence and collect new insights when reading these.

I remember doing this all manually when I was doing my master's degree in the UK. That was a very terrible and tedious experience partially because I've ADHD

So my question is, is it dishonest?

Because I can defend every claim in the review because I built the verification pipeline and reviewed manually each one. I arguably understand the literature better than if I had read it myself manually and highlighted all. But I know that many universities would consider any AI-generated text as academic misconduct.

I really don't quite understand the principle behind this position. Because if you outsource the task of proofreading, nobody would care. When you use Grammarly, the same thing. But if I use an LLM to create text from verified, structured, human-reviewed evidence — it might be considered dishonest.

In the last few months agent skills went from a niche Claude Code feature to something every major runtime supports. Anthropic has an official skills repo. OpenAI shipped skills in Codex with a built-in skill-creator. Karpathy talks about "everything is skill issue" and describes writing skills as curricula for agents [1]. The format is converging: a folder with a SKILL.md, optional scripts, optional reference files.

What changed is that the models got good enough to follow written instructions reliably. A skill is just a tested workflow in markdown that the agent reads and follows instead of improvising. You can also bundle scripts that the agent runs during the workflow, which covers what most people use lightweight MCP servers for, except the agent can read the script source and extend it.

Karpathy talks about an "economy of agents" and says we should stop writing HTML docs for humans and start writing markdown docs for agents [1]. Anthropic just shipped a skill-creator that benchmarks whether a skill still works after model updates. There are already tens of thousands of community skills on GitHub.

Distribution still feels early. Most useful skills are tiny. A markdown file, maybe one script. Useful enough to keep reusing, but not something anyone turns into a proper GitHub repo with a README and install instructions. So they stay on one machine.

I have been writing skills for my own agents for a while and I keep running into this. The format works. Moving them between machines or handing one to someone else does not.

Curious if others are hitting the same wall or if there are approaches I am missing.

[1] https://www.youtube.com/watch?v=kwSVtQ7dziU (Karpathy on No Briars podcast, skills discussion around 1:03:40)