Long have a specific failure mode: something plausible gets said early, nobody challenges it, and forty exchanges later it is load-bearing fact. The session cannot catch this, because the session is the thing that believes it.
This materialises a reviewer with fresh context and an explicit bias to disprove, before the decision stands.
How it differs from a code review
A review is a verdict on a finished artifact. This is an in-flight posture — it fires while changing course is still cheap, on decisions rather than diffs.
It triggers on the things you’d expect to matter: production auth, security-sensitive logic, high-stakes migrations, irreversible operations, unfamiliar code.
Why it’s marked advanced
Not because it’s hard to run — it isn’t — but because it costs and wall-clock on every non-trivial decision, and a beginner working on low-stakes tasks will feel that as friction without seeing the payoff. Start with the free version of the same instinct: review in a fresh session and be brutally honest. Graduate to this when you’re shipping something you’d have to explain if it broke.
Worth knowing
- Installing gives you 25 skills.
source-driven-developmenton this page is in the same bundle. - SSH clone errors on a managed laptop? Use the marketplace form:
/plugin marketplace add https://github.com/addyosmani/agent-skills.git.
Access
Public — no access request needed. MIT licensed.