RE: https://mastodon.social/@mgifford/116106634552309135
(I asked Mike on LinkedIn.)
Has anyone tried this, measured outcomes?
Asking because:
https://arxiv.org/abs/2602.11988
It found AGENTS.md “files tend to reduce task success rates compared to providing no repository context, while also increasing inference cost by over 20%.”
Wondering if someone with more smarts than I has tried applying the methodology in the study, can identify if / how it applies.
I have seen this problem repeatedly when talking to developers, maintainers, and accessibility practitioners: projects often lack a clear, living statement of how accessibility is handled. That gap is why I drafted ACCESSIBILITY.md guidance: https://lnkd.in/gpRZqKfS Why should projects have an ACCESSIBILITY.md? Because it makes a project’s accessibility commitments explicit, discoverable, and actionable for contributors and users. How can it help both humans and machines? For humans, it clarifies expectations, processes, and responsibilities. For machines, having a consistent, structured file enables automation, tooling, and compliance scanning to integrate accessibility into workflows rather than leave it to chance. This is a community effort. The initial draft was created with AI under my direction; I shaped the intent, structure, and editing. I want practical feedback. What’s missing? What should be clearer? What formats or examples would make this genuinely useful for real projects?
@aardrian The methodology doesn't lend to practical application per-say (e.g., it's an experimental design) and no discussion section (which is where you propose potential practical and theoretical approaches to follow-up with).
Taking study as-is, devs should manually write/test their context files before use; otherwise risking increased failures and costs. But study needs additional robustness; other contexts would have to be further vetted (which they do note; one size doesn't fit all).
Adrian Roselli, pH0
Justin Ribeiro, Ph.D.