Detailed Analysis
A Reddit user operating under the handle bluemozilla has released "i-have-dyslexia," an open-source Agent Skill designed to reformat Claude Code's output into a more scannable, dyslexia-friendly structure. The tool is a single SKILL.md file that instructs Claude to follow specific formatting rules: one idea per sentence with no nested clauses, short plain-language word choices except where technical terms are necessary, small sections broken up by clear headings, and lists in place of dense paragraphs whenever steps or sequential information are involved. The creator built it after finding that Claude's responses—particularly following the Opus 5 release—had become long, text-heavy walls of prose that were difficult to parse, especially for someone with dyslexia who relies on visual landmarks and scannable structure rather than linear reading.
The project is explicitly framed as a response to an existing tool, ayghri/i-have-adhd, which addressed a related but distinct problem. That ADHD-focused skill was built to help users maintain focus while reading Claude's output, but the creator of i-have-dyslexia noted it didn't solve the separate challenge of comprehension for dyslexic readers, who often need information broken into short, non-nested sentences and clear headings rather than simply shorter or more focused text. This distinction matters because it highlights that accessibility needs are not monolithic—different cognitive and neurological differences require different interventions, even when the surface complaint ("this is hard to read") sounds similar.
The technical approach is notable for its simplicity: rather than building a wrapper application, browser extension, or client-side post-processor, the creator leveraged Anthropic's Agent Skills framework to directly shape how Claude generates its output at the source. This is emblematic of a broader shift in how developers are extending and customizing AI assistant behavior—instead of manipulating outputs after the fact, skills and system-level instructions let users bake formatting, tone, and structural preferences directly into the model's generation process. Because Agent Skills are lightweight, portable, and shareable (a single markdown file hosted on GitHub), they lower the barrier for non-engineers or accessibility-focused users to contribute meaningful tooling to the Claude ecosystem without needing deep technical infrastructure.
This development also reflects a growing grassroots movement around accessibility customization for AI coding assistants. As tools like Claude Code become central to how developers interact with codebases, the default verbosity and prose style of model outputs increasingly function as a de facto interface design choice—one that doesn't work equally well for all users. Community-built skills addressing ADHD, dyslexia, and likely other cognitive accessibility needs suggest that Anthropic's extensibility model is being used organically to patch usability gaps that a single default output style cannot address. This bottom-up accessibility work parallels trends seen in other software ecosystems (browser extensions, screen readers, dyslexia-friendly fonts) and signals that as AI assistants become primary daily-use tools, demand for personalized, cognitively-adapted interaction styles will likely continue to grow, potentially pressuring companies like Anthropic to build more of this configurability into core products rather than leaving it entirely to community skills.
Read original article →