KYC Rules Triage
Turns parsed onboarding records into a risk rating, document gaps, rule outcomes, and a routing decision.
The skill separates the trusted firm rules grid from untrusted applicant documents and explicitly says it scores/routes rather than approves. It provides some data-flow transparency. Points are deducted because least privilege, user confirmation, sensitive KYC-data handling, MCP dependency security, rollback, and rule-version verification are unspecified.
The steps, factors, document statuses, per-rule outcomes, and JSON disposition contract are reasonably clear, and each outcome must cite a rule. Points are deducted because unavailable inputs, missing fields, malformed data, unavailable screening MCPs, and conflicting results have no diagnosable failure procedure; no execution evidence is available in this static review.
The audience, trigger sequence after kyc-doc-parse, and non-approval boundary are clear for KYC/AML evaluation. Points are deducted because non-fit scenarios, exclusion triggers, rules-grid format requirements, mainland-China reachability, and Chinese-language support are not stated.
The skill uses a readable stepwise structure with input guidance, a factor table, an output example, repository-level Apache-2.0 licensing, and plugin-validation CI. Points are deducted because the skill itself lacks versioning, changelog, named maintenance responsibility, installation/dependency notes, FAQs, and troubleshooting guidance.
It directly addresses risk rating, document checks, rule-by-rule outcomes, and disposition routing, with an output structure usable by downstream workflows and an explicit prohibition on approval. Points are deducted because results depend on an unspecified firm rules grid, parser, and screening MCP, and no representative outputs or tests are statically verifiable; human review remains necessary.
The workflow, fields, and output contract in SKILL.md are auditable, and repository CI validates plugin structure. Points are deducted because there is no skill-specific key-path test suite, real execution record, third-party corroboration, or sample rules grid; the conclusion is based only on static source material.
- The rules grid, parsed record, and screening MCP are external prerequisites; safe-stop behavior and error feedback for missing, stale, or conflicting inputs are undefined.
- The skill handles highly sensitive KYC/AML data but specifies no redaction, access control, retention, audit logging, or cross-border transfer controls.
- Risk ratings and dispositions lack reproducible rule versions, thresholds, and sample tests; compliance personnel must review them and they must not be used directly as approval or rejection decisions.
What it does & when to use it
This skill takes the structured onboarding record produced by `kyc-doc-parse`, the firm’s KYC/AML rules grid, and sanctions/PEP/adverse-media screening results. It computes a risk rating, checks required documents against what was received, and outputs every applicable rule outcome with cited evidence fields. The rules grid is the trusted firm source; the applicant record is treated as untrusted input. It does not approve onboarding; it only triages and routes for human review.
Reads the structured applicant record, the firm rules grid, and screening results from the screening MCP. Scores jurisdiction, applicant type, ownership opacity, PEP exposure, sanctions/adverse-media hits, and source-of-funds clarity into a low/medium/high risk rating. Compares the required documents for the applicant type and risk level against documents_received and marks each item received, missing, or expired. Emits one row per applicable rule with rule id, rule text, outcome, and the fields that drove it, then returns the disposition JSON.
- A compliance analyst needs a traceable triage result for a new onboarding file that has already been parsed.
- An operations team wants to know whether a complex ownership structure needs more documents or an escalation.
- A screening team has a PEP or sanctions signal and needs a routing-only outcome, not an approval.
- A reviewer needs every conclusion tied back to a specific rule id and evidence field.
Pros & cons
- Combines risk scoring, document checks, and rule outcomes into one auditable output.
- Cites rule ids explicitly, which makes review and escalation easier.
- Treats the applicant record as untrusted input instead of accepting it at face value.
- It depends on the firm supplying an up-to-date rules grid and screening results.
- It is a triage tool, not an approval tool; even `clear` is only a routing outcome.
- It assumes you already have the structured output from `kyc-doc-parse`.
How to install
The skill lives at plugins/vertical-plugins/operations/skills/kyc-rules/SKILL.md inside the anthropics/financial-services collection. The README covers installing the collection or a vertical plugin bundle, but it does not document a skill-specific install command for this folder.
How to use
Run it after kyc-doc-parse, then provide the structured record, the firm’s rules grid, and sanctions/PEP/adverse-media results from the screening MCP. Ask for the risk rating, required-document check, rule outcomes with cited rule ids, and the JSON disposition fields (risk_rating, disposition, missing_documents, escalation_reasons, rule_outcomes).