This is a useful result, but it is important to note that this is not necessarily what people have in mind when they think of "LLMs generating skills." Having the LLM write down a skill representing the lessons from the struggle you just had to get something done is more typical (I hope) and quite different from what they're referring to.

I'm sure news outlets and popular social media accounts will use appropriate caution in reporting this, and nobody will misunderstand it.

> A common pitfall is for Claude to create skills and fill them up with generated information about how to complete a task. The problem with this is that the generated content is all content that's already inside Claude's probability space. Claude is effectively telling itself information that it already knows!

> Instead, Claude should strive to document in SKILL.md only information that:

> 1. Is outside of Claude's training data (information that Claude had to learn through research, experimentation, or experience) > 2. Is context specific (something that Claude knows now, but won't know later after its context window is cleared) > 3. Aligns future Claude with current Claude (information that will guide future Claude in acting how we want it to act)

> Claude should also avoid recording derived data. Lead a horse to water, don't teach it how to drink. If there's an easily available source that will tell Claude all it needs to know, point Claude at that source. If the information Claude needs can be trivially derived from information Claude already knows or has already been provided, don't provide the derived data.

For those interested the full skill is here: https://github.com/j-r-beckett/SpeedReader/blob/main/.claude...

+4.5pp for software engineering is suspiciously low compared to +51.9pp for healthcare. I suspect this reflects that frontier models already have strong SWE priors from training data, so skills add less marginal value. If true, skills become most valuable precisely in the domains where models are weakest — which is where you'd actually want to deploy agents in production. That's encouraging.

If you have the idea, more or less the implementation plan, let the LLM do the coding, you can end up with something maintainable and nice, it's basically up to you.

Strip away one layer, so you have the idea, but let the LLM come up with the implementation plan, then also the implementation, and things end up a lot less than ideal.

Remove another layer, let the LLM do it all, and it's all a mess.

When you create a skill for a particular model, you don't typically ask the model to create the skill based solely on its own latent knowledge. Otherwise, you'd expect the effect to be similar to telling the model 'make a plan before acting, make not mistakes'.

But that's what the paper's authors did!

When they say 'self-generated' they don't allow the model any tool access at all, not even web search.

It would be much more interesting if they had tested skills that were created in one of these ways:

A) The model interviews a human and then creates the skill, or

B) The model executes one or more deep research tasks in order to gather information, or

C) Some combo of the above.

However, I've found them to be useful for capturing instructions on how to use other tools (e.g. hints on how to use command-line tools or APIs). I treat them like mini CLAUDE.mds that are specific only to certain workflows.

When Claude isn't able to use a Skill well, I ask it to reflect on why, and update the Skill to clarify, adding or removing detail as necessary.

With these Skills in place, the agent is able to do things it would really struggle with otherwise, having to consume a lot of tokens failing to use the tools and looking up documentation, etc.

I have seen some devs pull out absolutely bad guidance by introspecting the code with the LLM to define "best practices" and docs because it introduces its own encoded biases in there. The devs are so lazy that they can't be bothered to simply type the bullet points that define "good".

One example is that we had some extracted snippet for C#/.NET that was sprinkling in `ConfigureAwait(false)` which should not be in application code and generally not needed for ASP.NET. But the coding agent saw some code that looked like "library" code and decided to apply it and then someone ran the LLM against that and pulled out "best practices" and placed them into the repo and started to pollute the rest of the context.

I caught this when I found the code in a PR and then found the source and zeroed it out. We've also had to untangle some egregious use of `Task.Run` (again, not best practice in C# and you really want to know what you're doing with it).

At the end of it, we are building a new system that is meant to compose and serve curated, best practice guidance to coding agents to get better consistency and quality. The usage of self-generated skills and knowledge seems like those experiments where people feed in an image and ask the LLM to give back the image without changing it. After n cycles, it is invariably deeply mutated from the original.

Agentic coding is the future, but people have not yet adapted. We went from punch cards to assembly to FORTRAN to C to JavaScript; each step adding more abstractions. The next abstraction is Markdown and I think that teams that invest their time in writing and curating markdown will create better guardrails within which agents can operate without sacrificing quality, security, performance, maintainability, and other non-functional aspects of software system.

Web3 and JavaScript frameworks never had the nerd-sniping power of the AI ecosystem. I'm not denying the usefulness and potential of the space, and the achievements of its current champions, but the degree with which it has consumed discussion and productivity in the tech space is worrying.

This article would be wildly interesting with the opposite headline, but instead it simply states what many of us would assume based on experience.

Reality doesn't force us to choose between skill or no skill, reality often doesn't give us a choice. You can either make a skill for your company's proprietary system or your model has to figure it out from scratch every time by searching wikis or reading code. If you use it right, skills are a compression mechanism. Instead of the process meaning your model needs to get all of theses files dynamically, it can simply statically run.

To steel-man the paper. It is worth looking at whether you should try to code something up first or try a skill first. And it may well be valid to say try first and if you can't work it out in 5 mins, install a skill. But there's a meta point of skills as software (where you deduplicate the effort of solving regressions).

For a reductio ad absurdum, If self-generated skills with no additional context _didn't_ eventually level off in performance, then we could reach AGI by making one big skill that keeps growing and solving harder and harder tasks, including improving the capability of its own skill-builder skill, all without embedding any signals from the environment or needing to interface with the real world at all.

If you want to say what you think is important about an article, that's fine, but do it by adding a comment to the thread. Then your view will be on a level playing field with everyone else's: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...

I think that most of the adoption around Agent Skills would have a focus on ease of use, standarization and context management and not correctness.

My own thoughts on how to approach skill building target people who are adopting LLM development now more than ever although this was definitely possible (in a non standard way before) [1]

[1] https://alexhans.github.io/posts/series/evals/building-agent...

But it seems pretty surprising to me. The training corpus contains so much information and the models operate at the level of… a bright novice. It seems like there obviously ought to be more insights to derive from looking harder at aspects of the corpus.

Why isn’t this considered astonishing?

Healthcare +51.9pp. Manufacturing +41.9pp. Software Engineering +4.5pp.

The domains where models have the weakest priors from pretraining benefit the most from external procedural knowledge. That's not surprising on its own, but there's an implication I haven't seen anyone raise: these are exactly the enterprise domains where that procedural knowledge is most proprietary and most dangerous to lose between sessions.

The paper's entire architecture is single-player. A SKILL.md sits in a directory, one agent reads it, session ends. When Agent A at a bank figures out the right approach to parsing 13F filings (0% to 75% with the right skill in this paper), that knowledge dies with the context window. Agent B starts from scratch.

We're building shared memory infrastructure for agents at Memco (https://memco.ai) and this paper maps directly to what our enterprise design partners keep telling us — the problem isn't writing skills, it's that procedural knowledge doesn't compound across agents, sessions, or teams. The paper even shows 2-3 focused skills outperform comprehensive docs, which is a retrieval problem masquerading as an authoring problem.

The question this paper should be asking isn't "can agents write their own skills" — it's "what infrastructure makes skills accumulate and transfer?" Static files in a folder is the wrong primitive for that.

Skills are most valuable precisely where models are weakest - domains with less training data, more proprietary knowledge, or specialized workflows. SWE is heavily represented in training data; healthcare is not. This is exactly what you would predict if skills encode what the model genuinely does not know, rather than regurgitate what it already does.

Building an agent OS (OpenClaw), we see this pattern constantly. Skills that move the needle are never 'here is how Python works' - the model already knows that. The ones that matter encode system-specific quirks, environment constraints, or hard-won lessons from real failures. colonCapitalDee shared a great rule above: only encode what is (1) outside the model training data, (2) context-specific to your environment, or (3) alignment guidance for future sessions. Everything else is regurgitation.

The paper tests pre-task self-generation with no external input - useless indeed. The interesting untested condition: skills generated through actual execution with real feedback, in domains with sparse training coverage. That is where +51.9pp starts to look like a floor, not a ceiling.

I have systemized and automated businesses for a long time before LLMs came out, which generally wasn't very popular.

It is really weird to see everyone get excited about this kind of automation and then try to jump to the end points with something that's non-deterministic and wonder why it doesn't work like every other computer they've used (all or none).

Agents can self generate skills, maybe not effortlessly, or with psychic skills of reading between the lines (special exception for Claude), it's also about the framework and scaffolding in which to create skills that work, and what can be brought back to the "self-generation".

Without experience in creating computer skills in general, attempts for self-generating agent skills is kind of trying to use AI to autocomplete a sentence and then not like how it went. To a fair degree it can be lined up to improve considerably.

Right now there seems to be a 6-12 month lag between studies like these and it being shared/reported in the wild.

Too often, they are researching something reported in the wild and trying to study it, and it very well may work for some cases, but not all cases, and the research kind of entirely misses it.

With AI, it's incredibly important to follow show and not tell.

Sharing this from genuine curiousity if this resonates with anyone, and if so, how/where.

1. only information and instructions on how to answer 2. some defined actions (run specific cli commands for specific tasks, use this api with those parameters) 3. skills including scripts

1 seems to be of limited use

2 and 3 can save the agent quite some time for finding a solution. And once the agent found a programmatic solution to a specific problem, they can store this information in a skill

Chaos Congress talk on this from a couple months ago, jump to the coding loops part: https://media.ccc.de/v/39c3-breaking-bots-cheating-at-blue-t... . The talk focuses mostly on MCPs, but we now use the same flow for Skills.

This kind of experience makes me more hesitant to take on plugin and skill repos lacking evals or equivalent proving measurable quality over what the LLM knows and harness can handle. Generally a small number of things end up mattering majorly, but they end up being pivotal to get right, and the rest is a death by a thousand cuts.

Just asking a model "how good is this skill?" may or may not work, possibly the next laziest thing you could do - that's still "for cheap" - is asking the model to make a quiz for itself, and have it take the quiz with and without access to the skill, then see how the skill improved it. But there's still many problems with that approach. But would it be useful enough to work well enough much of the time for just heuristically estimating the quality of a skill?

I.e. your use of skills resulting in differences of up to ~52 percentage points (or negative percentage points) in improvements (or degradations) in your percentage pass rate is a decent first-pass measure of the importance of skills here.

FWIW I didn't read the paper and am judging it based on its title, which I think is fair because "self-generated agent skills" is a pretty loose definition.

It's fairly common we notice these types of threads where one thing is being postulated and then there's comments upon comments of doer's showing what they have done.

Claude's training data is the internet. The internet is full of Express tutorials that use app.use(cors()) with no origin restriction. Stack Overflow answers that store JWTs in localStorage, etc.

Claude's probability space isn't a clean hierarchy of "best to worst." It's a weighted distribution shaped by frequency in training data.

So even though it "knows" stuff, it doesn't necessarily know what you want, or what a professional in production environment do.

Unless I'm missing something?

I conjecture that after some years of LLMs reading a SharePoint site, producing summaries, then summaries of those summaries, etc... We will end up with a grotesque slurry.

At some point, fresh human input is needed to inject something meaningful into the process.

I have actually found something close to the opposite. I work on a large codebase and I often use the LLM to generate artifacts before performing the task (for complex tasks). I use a prompt to say "go explore this area if the code and write about it". It documents concepts and has pointers to specific code. Then a fresh session can use that without reading the stuff that doesn't matter. It uses more tokens overall, but includes important details that can get totally missed when you just let it go.

Google's Aletheia works like this, and instead of degrading it keeps getting better. I get what you're trying to say, though. The less world knowledge you provide the LLM, which it otherwise lacks, the worse its outputs will be.

(This also suggests that you should expect them to generally be bad at judging novel self-generated prompts/skills - if they could judge those, they would already be using them! There is a generator-verifier gap, but it is already exploited heavily during post-training and not much low-hanging fruit left there.)

I don't completely disagree (I've argued the same point myself). But one critical difference between the LLM layer and all of those others you listed, is that LLMs are non-deterministic and all those other layers are. I'm not sure how that changes the dynamic, but surely it does.

I chose a title that directly reflects an interesting finding - something that offers substantial insight to the community. I think the rule should be applied with some nuance; in this case, being explicit is a net positive.

I have no interest in linkbait, and I hope that's evident from my previous submissions