KYC Intake Parser
Turn onboarding packets into structured KYC data ready for screening.
The skill explicitly treats inputs as untrusted, forbids executing document instructions, following links, or opening embedded content, and limits the task to extraction, reducing prompt-injection and external-effect risks. Points are deducted because it does not specify least-privilege access, redaction, retention, transmission, user confirmation, rollback, conflict handling, or source attribution for sensitive KYC data.
The ordered steps, schema, null-for-missing rule, and no-guessing requirement make the happy path reasonably clear. Points are deducted for no diagnostic handling of corrupt files, conflicting fields, invalid dates, duplicate documents, or unparseable input; no skill-specific tests are shown, so the static ceiling is applied conservatively.
The scenario, document categories, KYC fields, JSON output, and pre-rules-engine gap checks are clearly described. Points are deducted because non-fit boundaries, trigger conditions, language support, regional compliance variation, and adaptation across document formats or deployment environments are not defined.
The documentation uses progressive steps and includes a document taxonomy, JSON template, and safety boundary; repository context supplies installation guidance, an Apache-2.0 license, version-pinned CI validation, and contribution guidance. Points are deducted because the skill lacks its own version, changelog, named maintainer, update path, FAQ, dependency notes, and substantial troubleshooting or examples.
The skill can organize onboarding materials into identity, ownership, control, source-of-funds, tax, and document-inventory fields, producing structured data intended for the rules engine; the core value is clear. Points are deducted because no representative output is verified and there is no treatment of field validation, provenance locations, contradictions, privacy protection, or complex entity structures, leaving meaningful manual review.
The pinned revision, repository README, Apache-2.0 license, and commit-level CI workflows provide some traceability; CI includes plugin validation and secret scanning. Points are deducted because there is no evidence of key-path test coverage, sample input/output fixtures, third-party execution evidence, or independent corroboration, so the static score remains at or below 5.
- Use this only for document extraction; its output must not be treated as a KYC approval, rejection, or rules-engine determination.
- Identity, ownership, PEP, and source-of-funds fields are sensitive; deployment should add access control, redaction, retention, encryption, audit, and human-review requirements.
- The skill does not define standard error output or escalation for missing, conflicting, expired, malformed, or unreadable materials.
- There is no evidence of Chinese-language field handling or reachability and fit for mainland-China network environments.
What does this skill do, and when should you use it?
This skill parses investor or client onboarding packets and extracts identity, ownership, control, source-of-funds, tax, and document-inventory data. It inventories the received documents, produces one structured JSON record, and uses null when information is not found. It then flags obvious missing or expired materials for the downstream KYC rules engine. It performs extraction and gap identification, not final screening decisions or onboarding approval.
Reads applicant-supplied onboarding documents and inventories them by identity, entity formation, ownership and control, address, source of funds or wealth, and tax categories. Extracts applicant type, legal name, birth or formation date, nationality or jurisdiction, address, identity documents, beneficial owners, controllers, source of funds, PEP declaration, tax forms, and received-document metadata into one JSON record. Flags plainly missing or expired items, including expired identification, address proof older than three months, or a missing ownership chart for an entity.
- A compliance operations professional receives an individual onboarding packet and needs a structured inventory of identity, address, and tax documents.
- A fund or wealth-management team receives entity materials and needs to capture beneficial owners, controllers, and formation documents.
- A KYC analyst must convert unstructured onboarding materials into a consistent JSON input for a rules engine.
- An operations team wants a pre-handoff check for missing or expired identification, address proof, or ownership documentation.
What are this skill's strengths and limitations?
- Covers core KYC intake fields across identity, ownership, control, source of funds, PEP declarations, and tax forms.
- Requires null for unavailable fields, helping avoid unsupported guesses.
- Separates document inventory, structured extraction, and obvious-gap checks into explicit steps.
- Treats applicant-provided content as untrusted and prohibits executing embedded instructions or links.
- The source provides no test suite, accuracy metrics, or defined document-format coverage.
- It identifies intake gaps but does not run the KYC rules engine or make compliance or onboarding decisions.
- OCR, batch processing, external database verification, and automated network access are not documented.
- A standalone installation command for the leaf skill is not documented.
How do you install this skill?
The skill is located at plugins/agent-plugins/kyc-screener/skills/kyc-doc-parse/. The README documents adding https://github.com/anthropics/financial-services through Cowork Settings → Plugins → Add plugin and selecting the relevant agent or vertical, or uploading a zip of a directory under plugins/. It does not document a standalone installation command for this leaf skill.
How do you use this skill?
Provide the investor or client onboarding materials to a client that has loaded the skill and ask it to inventory the packet, produce the KYC JSON record, and list obvious gaps. Example: “Parse this client onboarding packet using kyc-doc-parse, return one JSON record, and list missing or expired materials separately.” The source does not specify an exact trigger syntax.