📜 introducing ossrules.md - learn how leading OSS projects use agent rule files - discover skill files that projects actually use - level up your own projects and workflows I've personally learned a lot just exploring

Looking for skill file ideas? Search 100 repos worth in seconds Below I quickly find some video generation skills

One useful filter when borrowing a rule: what mistake was it written to prevent? Reproduce that mistake in your own repo before adopting the instruction. A rule that protects one project can add noise in another with different tests, tools or conventions.

maybe we git blame lines so you can see the history of why it changed

Absent a skill aggregator that incorporates eval benchmarks into the listings, I just don't know how to confidently differentiate useful skills vs. those that seem intuitively beneficial but actually regress performance.

It’s a vibes-based eval: “This successful project / these credible maintainers use this so it’s probably not junk” But fwiw I’m not copying skills verbatim, mostly fishing for ideas to borrow

I'm sure there's some signal in certain patterns spreading memetically, but prominent projects/developers are a lot less rigorous about vetting these things than most people assume. At the moment evals/benchmarking reminds me of the quote about teenage sex, i.e. “everyone talks about it, no one really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.”

Like, there is stuff in the agent I work on that I'm almost certain other people have copied, on the assumption that we did at least some degree of quantitative validation; and in reality there are things that snuck in entirely on the basis of single engineer feeling good vibes at 2am in a mild state of AI psychosis.

Maybe I'm telling on myself here but I don't think this is a unique phenomenon.

I just really want a skill aggregator that automates benchmarking and reports metrics on at least a few permutations of model/harness/skills. But probably the real answer is we need tools to make benchmarking your unique setup against representative tasks as easy as possible.

I think you're a 0.1%er and I made this for the rest of us 😅 I mostly use this for discovery, e.g. I didn't know I could've been using this the whole time

Oh no I think this is awesome, I was just hoping to nerdsnipe you into doing this for me.

Some issues on iOS safari: - can’t use “tap to scroll to top”, likely because a nested “overflow css” element is used for body content. Fix by pinning you ui instead of scrolling a div. - “back” loses UI state. Ie scroll down the list of projects, tap one, go back; scroll state lost. Just make I html (it’s just a list; no need for JavaScript to build the dom, which could be the issue)

oof, thought i'd fixed some of those - thank you for reporting!

made a cli/mcp tool for the site so my agents can use it:

nice - I also added this prompt to each index to help agent discovery I debated doing an MCP tool for this but, I feel just throwing a prompt in the mix is faster/easier

Woahh this is very cool.

Cool! Let’s add @ApacheOpenDAL’s!

@ApacheOpenDAL If you can open an issue I’ll get on it

Laravel/PHP next? 👀

It’s meant to be a mix, I chose some but Astra chose most If you have some good example repos I’ll take a look

this is so cool ben thanks for sharing <3

np, thanks for the support!

glad to help anytime ❤️‍🔥

I kept wishing OSS repos would publish the unspoken rules in one place instead of burying them across Contributing, Issues, and tribal memory. @bentlegen on ossrules.md (how leading OSS projects use it) is what I'm parking next to the next open agent harness README, so newcomers get the review bar and the non-goals without a scavenger hunt. Notes went into the open-source agent craft shelf.

@kentcdodds Nice idea!

@kentcdodds Thanks! If you have any feedback, would love to hear

Will definitely try this, thanks for sharing. I thought they didn't have those.

Curious how many of these rule files are actually maintained vs. written once and left to rot. Agent rules drift fast once the underlying tooling changes, and stale ones are worse than none.

@sjmaple see this