Seven stars. Zero gained in the last week. If you're scanning SkillsMP for what's hot right now, blast-radius isn't it. But that's actually fine — and I'll explain why that's not a dealbreaker, even though it should give you pause.
What this skill actually does
blast-radius is a review skill built around one uncomfortable question: "What does this change break somewhere else?" Not what it does — that's how. Not why it was written that way — that's why. It specifically targets the invisible blast zone that a diff never shows you.
The author, Aqua-123, built this as part of a larger 48-skill suite called pstack-for-codex, which is Codex-native. That matters more than it sounds, and I'll get to it. But the skill itself is genuinely interesting regardless of platform.
The core philosophy is deceptively simple: stop writing up blast radius analysis that sounds convincing and start proving the one fact that makes your change safe. The skill explicitly calls out the trap — a well-written analysis reads as correct whether or not it is. Words are where you start, not what you ship.
Why this matters more than you think
Every developer has written a "this should be fine" review at 11pm on a Friday. You look at a diff, you trace the obvious callers, you nod, and you merge. The skill's author names the exact failure mode: grep can find callers in a second. What grep can't do is tell you that the JSON response from an API three hops downstream now has a different field, or that a feature flag you didn't check gates the entire code path, or that a Solid vs React lifecycle difference means your teardown runs at the wrong time.
This fills a real gap. Most review tools and skills focus on what changed. blast-radius focuses on what the change unmasks. It's the difference between reading a map and actually walking the terrain.
What I found compelling
Five things stand out from the SKILL.md:
-
The "one fact it's safe because of" heuristic. Most scary-looking changes are safe because of a single fact. The skill forces you to find that fact first instead of drowning in a list of maybes. This is genuinely smart prioritization.
-
The confidence ladder. Each safety claim gets rated from "you said so" (worthless) to "you reproduced it in the running app." Any fact that can't reach step 4 — a script or test that calls the real code — gets explicitly marked as unproven. No hand-waving allowed.
-
The "look where grep stops" directive. The skill explicitly calls out what symbol search misses: JSON shapes, wire formats, DB columns, feature flags, code three hops downstream, and timing concerns like microtasks and unmount events. This is the part most review skills skip entirely.
-
The arena integration. For bigger changes, you run it across multiple models and merge answers. Different models catch different real bugs. That's a pragmatic use of multi-model evaluation that most skill authors only dream about.
-
The output format. The writeback template is specific: what changed, the one safety fact (with proof or "unproven"), only real risks with file:line citations, what you cleared, and the cheapest test that catches the real bug. No filler. No padding.
Who should install this
If you regularly review diffs you don't fully trust and you want a structured way to pressure-test your conclusions, this skill gives you a real framework. It's especially useful if you work in codebases with deep dependency chains, complex state, or multi-language systems where the blast zone is genuinely hard to map.
If you're the kind of reviewer who already runs targeted repro scripts before signing off, you'll find this validates what you already do — and gives you a checklist for the parts you might skip.
Who should skip it
If you're looking for a quick diff scanner or an automated PR reviewer, this isn't it. It's a deliberative skill that expects you to read source code, write scripts, and run tests. It's also worth noting — and this is the part that gives me pause — the skill is built for Codex, not Claude Code. The README explicitly instructs Codex CLI installation. While the SKILL.md format is portable, the ecosystem references, the poteto-mode delegation layer, and the companion skills (why, how, arena, unslop) are all Codex-native. If you're a Claude Code user, you're installing a skill that was designed for a different runtime.
Concerns and limitations
Let me be direct. The skill is thoughtful, but the low star count and zero recent growth suggest the community hasn't validated it yet. I can't find public reviews or usage reports outside the author's own repo. The methodology is sound — it reads like someone who's been burned by exactly the failure modes it describes — but sound methodology and proven tooling are different things.
The Codex lock-in is the biggest practical concern. The skill references ../poteto-mode/references/codex-agent-runtime.md, which means it expects a Codex environment. Porting this to Claude Code would require replacing or reimplementing the delegation layer and at least two companion skills (why and unslop) to get the full experience. You could use blast-radius standalone, but you'd lose the orchestration that makes it work as designed.
There's also the question of scope creep. The skill asks you to write scripts, run tests, reproduce in the running app, and potentially run multi-model arenas. That's a lot of work for a "review" skill. It's honest about this — it tells you to get as far down the confidence ladder as is cheap — but "cheap" is relative, and for a five-line config change, this might be overkill.
The verdict
blast-radius is the most intellectually honest review skill I've read this year. Its methodology — find the one fact, prove it or mark it unproven, don't round up — is exactly how I'd want a review to work if I were the one merging. The execution is solid.
But it's a Codex skill wearing a portable SKILL.md. If you're on Codex, install it. If you're on Claude Code, the ideas are worth stealing, but the skill itself will fight your runtime at every turn. Seven stars might be an accurate reflection of that friction — people tried it, found the platform mismatch, and moved on.
The thinking is worth more than the packaging. Just know what you're buying.
Links
SkillsMP page: https://skillsmp.com/creators/aqua-123/pstack-for-codex/skills-blast-radius GitHub: https://github.com/Aqua-123/pstack-for-codex/tree/main/skills/blast-radius