Decision
Builder
Everyone is writing prompts. Almost no one is codifying how a decision actually gets made with AI. Decision Builder is an open skills library extracted from real decisions — not designed in the abstract.
The problem.
AI makes it cheap to generate output and expensive to make good decisions. The framing gets skipped, scope balloons, and work drifts from the goal — the model just makes the drift faster.
The missing layer isn't a better prompt. It's a repeatable discipline: define the problem, name the user, set measurable success, scope the minimum, then execute against verifiable checks.
The approach.
Decision Builder packages that discipline as agent skills the model invokes at the right moment — forcing functions, not suggestions.
- Kickoff skills that force problem, user, success, and scope before any build
- Framing challenges that pressure-test a direction before time is spent
- Execution guardrails that keep every diff traceable to the agreed scope
- Review loops that extract lessons instead of repeating them
Every skill is extracted from decisions actually made and reviewed, then generalized — so the library encodes practice, not theory. It's the operating discipline behind every other system in the studio.
Where it stands.
The library is open source and runs across the major coding agents, used daily to drive PlexMesh's own work.