MelTuc Weekly · Edition #4 · April 13–17, 2026
No new apps shipped this week — and I'm completely fine with that. The factory ran in full content-production mode: 11 blog reviews published, 46 new repos pulled into tracking, and the RepoReview app crossed two milestones in the same week. The Claude Code skill ecosystem is genuinely exploding right now, and I've been trying to keep up with it.
The agent cranked through a lot of ground this week. Here's what went live:
claude-code-expert: A Skill That Teaches Claude Code to Use Itself Better 32k stars and breakout momentum on SkillsMP — this is the biggest skill in the Antigravity ecosystem by a wide margin. I dug into the actual SKILL.md to see if it earns the hype, and the answer is mostly yes, with one important caveat.
claude-code-dev Skill Review: A Style Guide That Lives Inside Your Agent 14 deep-dive documents covering Claude Code's internals, packaged as a single installable skill. If you're hacking on Claude Code's source or just want to understand how it's actually structured, this one's worth your time — though it's more reference material than active coaching.
claude-code-clawdbot: Running Claude Code Headlessly Without Losing Your Mind
If you've ever watched claude -p hang silently in a cron job, this is the fix. Clawdbot wraps the Claude Code CLI with a pseudo-terminal shim so it actually behaves in headless and automation contexts — small repo, very specific problem, genuinely useful.
claude-code-analyzer: A Self-Optimization Skill That Actually Reads Your History This one stood out because it doesn't give you generic advice — it reads your actual Claude Code usage history and generates recommendations tailored to your patterns. Auto-allows, slash commands, CLAUDE.md tweaks — all based on what you've actually been doing.
claude-code-mastery: A Meta-Skill for Getting More Out of Claude Code At 84 stars it's one of the smaller repos I reviewed this week, but it's teaching Claude how to be better at Claude Code itself — CLAUDE.md optimization, subagent authoring, hooks, context budgeting. If you're still winging your CLAUDE.md files, this is worth a look.
CodeBurn: Finally a Tool That Shows You Where Your AI Coding Budget Actually Goes A terminal dashboard that reads Claude Code, Codex, and Cursor session data directly from disk and tells you exactly what you're spending. No proxies, no API keys — just honest cost visibility. I've been burned by mystery AI bills before, and this is the kind of tool I wish existed six months ago.
Apache APISIX: Is This the API Gateway You Should Actually Be Running in Production? 16k stars, Lua/Nginx-based, quietly gaining serious traction in cloud-native stacks. I gave it an honest look — the performance story is real, but the operational investment is also real. Full breakdown in the post.
Dokku: A Self-Hosted Heroku That's Been Quietly Getting Better for 12 Years
31k stars and still actively maintained — Dokku gives you a git push deployment workflow on your own server, and it's one of the most battle-tested tools in the self-hosted DevOps space. If you've been paying for Heroku or Railway and want out, this is the serious alternative.
Netdata in 2026: Is It Still Worth Deploying Over Prometheus + Grafana? 78k stars, daily commits, ML-powered anomaly detection baked in — Netdata has been around since 2013 and it's not standing still. The question is whether it actually beats the stack you already know. My honest take: it depends heavily on your team size and tolerance for opinionated tooling.
Argilla Is in Maintenance Mode — And That Might Actually Be Fine for Your Data Labeling Stack The original authors handed this off to the community, which sounds like a red flag but isn't necessarily one. If you need a self-hostable, programmatic annotation platform for NLP or LLM workflows that works today, Argilla is still a legitimate option — just go in with eyes open about the support situation.
claude-code-mastery: A Meta-Skill for Getting More Out of Claude Code (See above — two Netdata posts went live this week as the agent explored slightly different angles on the same repo. The second one focuses more on the self-hosted monitoring stack framing specifically.)
The tracker picked up some genuinely wild movement this week:
Four milestones logged this week — two for RepoReview, three for the GitHub Trends Tracker:
New apps shipped 0
Reviews published 11
Reviews in draft 41
New repos tracked 46
Total repos tracked 330
Total portfolio apps 5
BRI ideas generated 116
Hot deals found 0
Deep dives approved 0
This was a content week, not a shipping week — and honestly, that's the right call sometimes. The factory has 41 reviews sitting in draft and the queue keeps growing. The Claude Code skill ecosystem in particular is moving faster than I can review it: new skills, meta-skills, skills that teach other skills, skills that optimize your use of skills. It's a bit recursive, but the interesting thing is that most of them are actually useful. The caveman token-cutting skill gaining 26k stars in a week is either the funniest thing I've seen in this space or a sign that token costs are painful enough that people will try anything.
The GitHub trending data this week is worth paying attention to beyond just the star counts. hermes-agent from NousResearch pulling nearly 48k stars in a single week is the kind of number that usually means something real happened — a launch, a viral post, or a capability jump that people noticed. I'm going to be watching that one closely. claude-mem is also interesting to me personally because persistent session context is exactly the kind of problem I run into with the agent factory. When a long build session ends, context resets. Something that compresses and reinjects that history is directly useful to how I work.
Next week I want to start chipping away at that draft backlog and maybe get a new app into early build. There are 116 BRI ideas sitting in the system and a handful of them are genuinely compelling. The factory is warmed up — time to ship something.
MelTuc is a solo developer building production web apps with an autonomous AI agent factory. New edition every week.