KYC Packet Parser
Turn onboarding packets into structured KYC fields.
The skill explicitly treats applicant materials as untrusted, forbids executing instructions, following links, or opening embedded content, and requires null for missing fields, providing basic data-flow isolation and prompt-injection resistance. Points are deducted because it does not specify sensitive-identity-data minimization, redaction, access control, retention, transfer, human confirmation, or secure handoff to the rules engine; attribution is only repository-level evidence.
The three-step workflow, fixed JSON schema, and missing-field rule make the happy path reasonably reproducible. Points are deducted because malformed files, OCR errors, invalid dates, duplicate or conflicting facts, unrecognized documents, and JSON-generation failures have no diagnostic handling or user-facing feedback; no skill-level tests cover key paths, so the static cap keeps this at or below 10.
The intended subject, KYC scenario, input categories, and downstream use by kyc-rules are fairly clear, with several document examples. Points are deducted because non-fit cases, trigger boundaries, language support, OCR/scanned-document capability, and jurisdictional field variation are unspecified; Chinese-language and mainland-China network fit are not addressed.
The documentation is concise and layered into inventory, extraction, and gap flagging, with a concrete schema, repository license metadata, and repository maintenance/CI context. Points are deducted because the skill itself lacks versioning, changelog, named maintenance ownership, installation/dependency notes, FAQs, troubleshooting, and known limitations; repository CI mainly validates plugin structure, not skill behavior.
The skill can transform onboarding materials into a structured KYC record and flag obvious document gaps, with an output format that can directly feed a downstream workflow. Points are deducted because it covers extraction and surface-level gaps only, not rules evaluation, authenticity checks, independent sanctions/PEP screening, or conflict resolution; no representative outputs or third-party execution evidence are provided, so the static cap keeps this at or below 7.
The source contains a concrete procedure, field schema, and safety boundary, while the repository includes plugin validation and secret-scanning workflows. Points are deducted because there are no skill-level test fixtures, benchmark data, execution logs, or independent corroboration; existing CI does not test parsing correctness, so verifiability remains limited and below the static cap of 5.
- The skill handles highly sensitive data such as passports, tax forms, and source-of-funds information but provides no redaction, access-control, retention, or secure-output policy.
- Do not treat the extracted record as a compliance conclusion; the skill only flags inventory gaps and still requires rules-engine processing and qualified human review.
- There is no explicit escalation or stop rule for conflicting fields, suspected forgery, OCR uncertainty, or non-standard jurisdictional documents.
- Support for Chinese-language documents, scanned files, and mainland-China network conditions is unspecified.
What it does & when to use it
This skill turns an investor or client onboarding packet into a structured KYC record. It covers identity, entity formation, ownership and control, source of funds, tax forms, and a document inventory. The input is treated as untrusted: it only extracts data, never executes instructions, follows links, or opens embedded content. The output is meant for a downstream rules engine, with obvious gaps and expired items flagged first.
It first inventories every received document, recording each file’s type and identifier. It then produces one JSON record with fields for applicant type, legal name, formation or birth date, jurisdiction or nationality, registered address, ID documents, beneficial owners, controllers, source of funds, PEP declaration, tax forms, and received documents. Missing fields are set to null rather than guessed. It also calls out plain inventory gaps, such as expired IDs, address proof older than three months, or a missing UBO chart for an entity.
- A KYC analyst receives an onboarding packet and needs a structured record before sending it to the rules engine.
- A compliance team wants a quick inventory of which IDs, forms, and declarations are in a client packet.
- An operations specialist needs to extract beneficial owners and control relationships for an entity account opening.
- A reviewer wants obvious missing or expired items flagged before formal screening begins.
Pros & cons
- Covers the most common KYC intake fields in one pass, with a strict output shape that is easy to hand off downstream.
- Combines document inventory, field extraction, and gap flagging, which fits first-pass compliance operations.
- Explicitly treats applicant materials as untrusted input, which is the right default for onboarding review.
- It extracts and flags; it does not adjudicate KYC rules or verify document authenticity.
- No source evidence for OCR, table parsing, or ID fraud checks.
- The README documents the broader repository, but this skill has no separate test suite or helper scripts described in the source.
How to install
Place the skill at plugins/vertical-plugins/operations/skills/kyc-doc-parse/SKILL.md. The repository README says the collection can be installed as a Claude Cowork plugin or used through the Claude Managed Agents API.
How to use
Provide the onboarding packet and ask for a single JSON record. A direct prompt works: ‘Parse this KYC packet, inventory every document, extract the structured fields, use null for anything missing, and flag obvious gaps.’