Every RolePlugin persona clears the same five-stage bar before it ships: research, spec review, two rounds of evals, adversarial review, and human sign-off.

Personas assist with real professional work, so we hold every capability skill to the same process, regardless of how small the change is. Here is what that process actually looks like. That consistency matters: a shortcut in one capability can undermine trust in the whole persona.

Research first

Before we write a single skill, we ground the persona’s domain knowledge in 8 or more research sources. This sets the vocabulary, the common workflows, and the edge cases a persona needs to recognize before it ever talks to a user. We also look at where a role commonly runs into judgment calls that don’t have one right answer, so the persona knows when to flag uncertainty instead of guessing.

Spec review

Each capability inside a persona starts as a written spec: what it does, what it needs from you, and what it hands back. That spec goes through review before anyone builds it, so we catch scope problems and missing edge cases early, while they are still cheap to fix.

Two rounds of evals

Once a capability is built, it runs through two full rounds of evals against realistic scenarios drawn from the research phase. The first round catches obvious gaps. The second checks that fixes from the first round did not introduce new ones. Both rounds run against the same scenario set, so we can tell whether a fix actually solved the underlying issue or just moved it somewhere else.

Adversarial review

After evals pass, a reviewer tries to break the persona on purpose: ambiguous requests, missing information, prompts designed to push it past its scope. The goal is to find the failure modes before a customer does.

Human sign-off

No persona or persona update ships without a human reviewing the final result and signing off. Automated checks catch a lot. They do not catch everything, and this is the last check before anything goes live.

Hard rules for liability-sensitive personas

Some personas work close to real liability: the Contract & Legal Reviewer, the GRC & Compliance Analyst, and the Financial Analyst. These carry explicit hard rules on top of the standard process. They will not issue legal or financial determinations, and they are built to support a licensed professional’s judgment, not substitute for it.

This is the same bar every persona in the RolePlugin catalog clears before you ever see it, and the same bar every update has to clear again.