Two thousand, five hundred ninety-three stars on SkillsMP and zero movement in the last week. That's an odd signal. Either this skill is quietly doing its job so well that nobody needs to talk about it, or it's too niche to matter outside one repo. After reading through the full SKILL.md and understanding what alexbelgium actually built, I'm leaning toward the first explanation — and that's exactly why I'm writing this.
What This Skill Actually Is
This isn't a prompt template or a code generator. The hassio-addon-workflow skill is a meta-workflow — a set of rules and stages that governs how an AI agent approaches Home Assistant add-on development tasks. It was built by alexbelgium, who maintains 143 add-ons with over 344,000 installs. That's not a hobbyist publishing a side project; that's someone with deep operational experience in a specific codebase writing down the process they actually follow.
The skill defines two paths: a light path for typo fixes, version bumps, and CHANGELOG edits, and a full loop for anything involving performance, diagnosis, or changing shipped defaults. The full loop has eight steps — scope, measure, plan, Codex review, implement, simplify, Codex review again, PR, resolve comments, verify in production, report. Every step has specific requirements, and the skill doesn't let you skip around.
Why It Matters
The gap this fills isn't about Home Assistant specifically. It's about the fact that most AI coding skills are undisciplined. They'll happily refactor your entire codebase, add abstractions you didn't ask for, and write 200 lines where 10 would do. This skill fights that instinct with concrete mechanisms.
The "proportionality question" is the best example. Before implementing anything, the skill forces you to answer in writing: What is the smallest thing that solves this? What would you delete from what I just proposed? It names the simpler alternative you rejected and requires justification. This is a direct countermeasure against the AI tendency to over-engineer.
The evidence-before-reasoning requirement is equally blunt. For RAM/CPU work, you run scripts/measure.sh with a 20-second sample. You don't reason about performance from source code — you measure the running add-on. The skill explicitly calls out that "reviewers hold you to the numbers." That's a culture shift, not just a procedural one.
Key Capabilities Worth Highlighting
First, the light vs. full loop split is genuinely smart engineering. A version bump doesn't need a Codex review. The skill knows that, and it saves expensive AI cycles for when they're actually warranted. Escalation is defined clearly — touch a default, add a new file, reveal a deeper problem, and you're on the full loop automatically.
Second, the mechanism ladder (levels 1-6) forces you to start at the simplest possible intervention. Config value first. Existing base image knob second. New init script only as a last resort. Each level up requires a reason that "survives being said out loud." "It felt cleaner" is explicitly called out as insufficient.
Third, Codex as an independent reviewer — not a co-pilot, not a rubber stamp. The skill asks Codex to argue against the plan, and it specifically instructs you to sketch the smaller alternative for Codex to challenge. This is the one question in the entire loop that pushes toward less code, and the skill treats it as uniquely valuable.
Fourth, the standing rule on simplicity is worth quoting directly: "120+ add-ons are maintained by one person: a homogeneous repo where every add-on solves a problem the same way is worth more than a locally nicer bespoke design." This is a philosophy disguised as a coding guideline. It prioritizes maintainability over elegance, and that trade-off is rarely articulated this clearly.
Who Should Install This
If you contribute to alexbelgium's hassio-addons repo, install this immediately. It's purpose-built for that codebase, and the scripts, templates, and conventions all point to it.
If you're a Home Assistant power user who modifies add-ons locally and wants a more disciplined approach to your changes, this is still worth studying — even if you don't use the repo-specific scripts, the workflow philosophy transfers.
If you're a developer who works on Docker-based services and wants a template for an evidence-driven AI workflow, skim the SKILL.md for ideas. The proportionality question alone is worth the read.
Who should skip this: developers working on unrelated projects. The skill is deeply tied to this repo's structure, scripts, and conventions. The scripts/measure.sh, .templates/ patterns, and scripts/preflight.sh all point to specific paths in the alexbelgium repository. You'd be adopting the philosophy without the scaffolding, and honestly, you'd get more value from writing your own workflow than from grafting this one onto a different codebase.
Concerns and Limitations
The biggest concern is lock-in. This skill is not portable by design. Every scripts/… and references/… path is relative to the skill root inside the repo. The non-negotiables — no Docker builds locally, work in /data not /tmp, never git stash under /data/claude — are all repo-specific constraints, not universal best practices.
The full loop requires Codex access, which means a paid subscription. The light path might work with Claude alone, but the independent review steps that make this skill distinctive are Codex-dependent. If you're on Claude Code only, you're getting roughly 40% of the value.
The skill is also dense. The SKILL.md is thorough to the point of being exhausting. There's a real learning curve, and the answer-style requirements (terse replies, no emoji, never compress uncertainty markers) are unusual enough that it'll take a few runs before the workflow feels natural rather than forced.
One more honest note: the trend status is listed as "Unknown" despite 2,593 stars. That might mean the skill is mature and stable, or it might mean it's not being actively promoted. Either way, it's worth noting that this isn't a skill that's riding a hype wave — it's been quietly accumulating stars for a reason.
Verdict
Install it if you touch this repo. Study it regardless. The proportionality question and the evidence-before-reasoning mandate are the most useful things I've read in any AI coding skill this year, and they're not Home Assistant-specific at all. The rest is scaffolding around those two ideas. Give it a 7/10 — brilliant philosophy, rigid implementation, and only useful if your repo matches the author's.
Links
SkillsMP: https://skillsmp.com/creators/alexbelgium/hassio-addons/claude-skills-hassio-addon-workflow GitHub: https://github.com/alexbelgium/hassio-addons/tree/master/.claude/skills/hassio-addon-workflow