Making things with AI.
I like making images, building projects, and tinkering with personal agents using AI. Here I share my experiments and original Skills, alongside software, Skills, and reading I recommend from other creators.
Skills
Original skills for AI coding agents, with instructions, source, and license included; recommended skills link to their original project.
Obsidian Markdown Skill: check the rendered note, not just the fileRecommendedOtherSteph Ango’s instructions for Obsidian-specific Markdown, including properties, wikilinks, embeds and callouts. Start with a test note and verify links and rendering. · Source: Obsidian Skills · kepano
Playwright CLI Skill: make browser checks reviewableRecommendedOtherMicrosoft’s browser-automation Skill for coding agents: inspect a snapshot, act on element references and retain evidence. Start in a dedicated test session and account for the reported attach limitation. · Source: Microsoft Playwright CLI
- Test Cases and Acceptance ValidationEngineeringDesign test cases based on confirmed requirements and real failure mechanisms, clearly distinguishing between case design, red test handoff, validation, and authorized fix closed loops.
- Closed-loop DevelopmentEngineeringCoordinate development, testing, defect fixes, regression, and review based on confirmed requirements and implementable designs. Assess delivery status using complete user tasks and evidence.
Software I recommend
Software worth a look from other creators, with credit and links to the original projects.
- Yazi: Browse, Preview, and Organize Files in the TerminalA terminal file manager with file preview, multi-tab, and customizable workflows. Additional search and media preview should be chosen in combination with auxiliary tools and actual terminal capabilities.
- Tad · A CSV and Parquet viewer with pivot tablesInspect CSV and Parquet files locally: filter rows, group data, build hierarchical pivots and export a filtered result. A viewer for analysis, rather than a spreadsheet editor.Created by Antony Courtney
- Bruno: keep API request collections beside your codeA desktop API client that saves request collections as local files. Review them in Git, start with an authorized read-only request, and check authentication separately for examples and ordinary requests.
Notes & recommendations
My notes on working with AI, alongside reading I recommend from other creators.
How to tell whether AI-written tests can actually fail
A green test proves nothing until it fails on wrong code. Break the guarded line on purpose, re-run, restore; use mutation testing to scale the check.
What fzf is for (and what it is not)
fzf selects one line from a list you already have. It is not a full-text content searcher. Learn plain selection before shell bindings.
A terminal workflow with fzf, lazygit and yazi
How fzf, Yazi and Lazygit fit into one terminal loop: find files, preview them, then review and stage Git changes. Based on each tool's official docs.
Recommended reading
Memory leaks: follow what keeps an object aliveA Google Chat case study makes retaining paths a useful starting point for investigating memory growth in long-lived web apps.Curated · Chrome for Developers · Vatan Aggarwal and Thierno Thiam · Nov 01, 2026
CrUX Soft Navigation Metrics: Looking at Each SPA Page Navigation SeparatelyAnnie Sullivan introduces experimental LCP, INP, and CLS metrics that include SPA route changes; changes in experimental scores do not mean the application itself has become faster.Curated · web.dev · Annie Sullivan · Oct 25, 2026