Adjudicating Screening Alerts With the Help of a Claude Skill
Amir Fadavi · 2026-05-19
Our mission is to make financial crime compliance more efficient by using AI for repeatable tasks in financial crime compliance domaine.
Here is an example of what we built recently:
A Claude skill that helps users adjudicate hits generated from a screening system. The Skill goes through a four-tier decision tree, consisting of more than ten rules, to determine whether a hit is a true positive, false positive, or needs to be escalated. This skill is build with a conservative approach and will escalate alerts when it is not confident if a hit is a true positive or false positive. Also, we built it to be deterministic (which means it will generate the same results when given the same input). Lastly, as we built it, we considered token burn rate by instructing it not to burn token if the information available is so limited that doing any work on the hit will only lead to inefficient use of tokens.
To add this skill to your Claude, go to Cowork tab > Customize > Skills > Click on + > Create a skill > Upload a Skill > Then drop the zip file you downloaded from Github.
If you have any questions, or if you want us to run this for you and create a reoccurring auditable log, send us an email to hello@sanctrust.com.