Blind Dev article
An AI assistant with skills: why I keep adding capabilities
Eight skills for everyday work and development: dates, documentation, updates, forms, code and accessibility.
An archival account from April 2026. The tools and results described refer to that period.
Four skills I use myself
By April 2026 I had spent about two months digging deeply into AI agents. Some things I found through acquaintances and communities; others I made for my own tasks.
Skills are files containing instructions and logic for an AI assistant. They help it do something genuinely useful rather than just talk.
plan-times
It is frustrating to say “the call is next Wednesday” and have AI confuse days, dates or deadlines. This skill gives the assistant a reliable reference for the current date and time so it can calculate days, weeks and deadlines correctly. It is useful for discussing plans, meetings and reminders.
getdocs
This downloads a project’s documentation so I can study it locally. The assistant can then work with files at hand rather than scattered fragments from the internet. It is particularly useful when documentation is extensive and spread across a website.
safe-update
I made this skill for updating OpenClaw after hearing about people whose installations broke after an update and took days to recover. It checks the current state and compatibility, creates a backup, updates, then checks that everything is alive. Updating becomes a calmer process rather than a lottery.
dashboard
This creates pages with sections, forms and tables. It produces HTML and JSON files: open the HTML, import the JSON, and work with them. I used it to make a more convenient post plan and, before that, a brief form as a test. I like being able to build something for myself quickly and pass it to someone else if needed.
In the original post I also suggested trying skills in ChatGPT, Cursor or VS Code: they would probably be able to use them too, except for safe-update. That was a suggestion, not a compatibility guarantee.
Four more skills for developers
AI lowers the entry barrier to development. Perhaps someone will use these skills to build a large project — maybe even one worth a billion 🙂.
coding
This is my foundation. My earlier setup used MiniMax for task definition and Codex for planning, development and testing. By the time of the post I was more often using ChatGPT Codex. The models and their roles can be changed.
I like the stages: task definition, planning, implementation and testing. They can run within a subagent without disrupting other processes.
accessibility-auditor-skill
I am blind and use a screen reader that speaks the interface. Many websites and apps still have poor accessibility. I made this skill to help an AI agent find those problems; coding or another tool can then help fix them.
I described it as suitable for mobile apps, desktop apps and websites. At that point I was still looking for volunteers to test it.
github-and-vercel-deploy
I did not feel like installing GitHub CLI and Vercel CLI, so I asked AI to make a skill using their APIs. It implemented it, removing my need for those extra programs. It is useful for publishing code to GitHub and deploying to Vercel without separate CLIs.
nvda-skill
My screen reader, NVDA, supports add-ons. I did not know Python, so I made a skill that helps create them with AI. This let me build things I needed, including Todoist for NVDA: an add-on for working with Todoist tasks in a convenient NVDA window.
Why this reminded me of Alice
Developers could once add skills to Alice, extending it beyond timers and music. That reminded me of modern AI-agent skills, although the comparison is not exact.
Alice’s skills were more limited, usually separate scenarios and functions. Modern skills in systems such as OpenClaw can form part of a whole workflow: writing in your style, analyzing data, helping with development and running multi-step tasks.
The underlying idea was already there: you do not have to use an assistant only as supplied. You can extend it for your own tasks. Those extensions have now become deeper and more useful in real work.