Runs one-off checks on a person with Didit, without a full verification flow. Use when the user says "sanctions check", "PEP check", "adverse media", "AML screen this name", "check this ID document", "proof of address", "validate against government databases", "compare these two faces", "face search", "liveness check", or "send a phone code" in their Didit account.
Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Didit Run Checks?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/didit-protocol-didit-run-checks)
---
name: didit-run-checks
description: Runs one-off checks on a person with Didit, without a full verification flow. Use when the user says "sanctions check", "PEP check", "adverse media", "AML screen this name", "check this ID document", "proof of address", "validate against government databases", "compare these two faces", "face search", "liveness check", or "send a phone code" in their Didit account.
---
# Run one-off checks
Follow the rules of didit-connect first: call `didit_context_get`, pick the application, and note whether it is sandbox or live.
## Before any check
- Every tool in this skill is a write. In a live application this can use your Didit credits (for example, each phone code sent is charged). Say so and wait for the person to agree before the first one. Sandbox applications are free to test in.
- When the person gives the checked person's id in the customer's system, pass it as `vendor_data` so the result links to that user. Only the AML, database, document, address and face tools declare `vendor_data`; the email and phone tools do not, so never send it to them.
## Checks that take text
**AML screening:** `didit_verify_aml`
- `full_name` (required). `entity_type` is `person` (default) or `company`.
- `date_of_birth` (YYYY-MM-DD) and `nationality` improve the match. Optional `document_number`.
- `include_adverse_media: true` adds adverse media. `include_ongoing_monitoring: true` enrols the entity for ongoing monitoring; ask before setting it.
- Report the hits by list type (sanctions, PEP, adverse media) with their match scores. A possible match is not a confirmed match: say so.
**Database validation:** `didit_verify_database`
- `issuing_state` (required, ISO 3166-1 alpha-3, for example `ESP`) selects the official sources.
- Add what the person has: `first_name`, `last_name`, `date_of_birth`, `document_number`, `document_type` (`P`, `ID`, `DL` and others), `personal_number`, `tax_number`.
- Optional `partial_match_action` and `no_match_action` (`DECLINE` or `NO_ACTION`).
## Checks that take files
On the hosted server, every file goes in a `<field>_base64` argument, as raw base64 or a data URL:
| Tool | File arguments |
|---|---|
| `didit_verify_id` | `front_image_base64` (and `back_image_base64`), optional `perform_document_liveness`, `minimum_age` |
| `didit_verify_poa` | `document_image_base64`, optional `expected_address`, `expected_country` (ISO alpha-2), `expected_first_name`, `expected_last_name` |
| `didit_verify_face_match` | `image_1_base64`, `image_2_base64` (order does not matter) |
| `didit_verify_face_search` | `image_base64` (searches every enrolled face of the application) |
| `didit_verify_passive_liveness` | `image_base64` (one selfie, no interaction) |
Use a file check only when your client can pass the real bytes of a file the person attached. Never ask the person to paste base64 text, and never make up image content.
If the client cannot pass the file, create a verification session instead (see didit-verify-people) with a workflow that has the right step, and share its link. The person being checked then uploads the document or takes the selfie themselves.
## Email and phone codes
These reach a real person. Confirm the address or number, and that the person being checked expects a code, before sending. The four tools take only the destination (and `code` for a check), plus the scope ids.
**Phone:**
1. `didit_verify_phone_send` with `phone_number` (with country code, for example `+34600000000`). Codes go by WhatsApp first and fall back to SMS. A code is valid for 5 minutes. The result has `request_id` and a `status` (`Success`, `Retry` or `Blocked` with a reason).
2. The person being checked reads the code and tells the operator. If the code should stay private to that person, offer a verification session with a `PHONE_VERIFICATION` step instead (see didit-verify-people), so they enter it themselves.
3. `didit_verify_phone_check` with `phone_number` and `code`. Each code allows 3 attempts. The status is `Approved`, `Failed` (with attempts left), `Declined` or `Expired or Not Found`.
In a sandbox application the phone send is mocked and the check approves any code. In a live application a phone send answers an error until the organization has credits, and repeated sends to one number can be rate limited. Both tools need a role that can create sessions.
**Email:** `didit_verify_email_send` and `didit_verify_email_check` (with `email`, and `code` for the check) are listed, but in this server build they call a route the server's own source marks as not existing, so expect them to fail. To confirm an email address, create a verification session whose workflow has an `EMAIL_VERIFICATION` step instead (see didit-verify-people) and share its link. If the person still wants the direct tool, try it once and report the error as returned; never claim a code was sent unless the result says so.
## Reading results
Lead with the outcome (approved, declined, needs review, match or no match), then the two or three facts that decided it. Keep personal data to what the person needs.
## What good output looks like
- One line naming the check, the application, and sandbox or live.
- The outcome first, then the key evidence, then any warning.
- For file checks the client could not pass, a verification link instead of a request for base64.
- A clear question before any billable check or any code sent to a real person.
## Do not
- Never ask for credentials (password, API key, one-time code to sign in). The verification code a checked person receives is not a credential of the operator.
- Never fall back to raw HTTP calls (curl/REST) for a console task.
- Never invent ids, names or image data.
- Never state prices. Link https://didit.me/pricing.