# Work from the native AI Skills

Give compatible agents task-specific workflows grounded in the same identity, proof and execution primitives.

Canonical: https://docs.bankeiz.com/ai-skills/
Reviewed: 2026-10-25; coordinated packages 129.0.0; ruleset 127.0.0.

## Use the coordinated release

```sh
npm install @receiz/ai-skills@129.0.0
```

Install alongside matching SDK and MCP packages in the integration project.

The Skills package supplies task workflows, manifests and prompts for supported identity, proof, ownership, subject and payment work. It uses the same primitives as SDK and MCP. A skill description is not a new issuer, verifier, permission grant or source of financial truth. Read its workflow and the actual installed APIs together.

## Begin with the task and source

1. **Choose the intended operation** — For example, inspect an Original, receive a supported Note or continue an admitted subject.
2. **Provide the actual source** — Use complete originals and the required identity/proof context, not only a screenshot or guessed ID.
3. **Inspect prerequisites** — Check exact package generation, host, grants and custody.
4. **Follow the native lifecycle** — Verify, prepare, review/authorize where required, execute and retain the accepted result.

## Keep instructions below authority

A prompt can explain a workflow or ask an agent to propose a next step. It cannot grant ownership, forge consent, make an unsupported host capable or replace a required verifier. If skill prose and installed package exports disagree, inspect the actual declared/source path and record the exact compatibility boundary. Do not create a missing API from a plausible name or use historical subject instructions for modern subject authority.

## Carry context deliberately

**Useful task context**

| Context | Handling |
| --- | --- |
| User goal | Natural-language outcome and intended scope. |
| Source references | Exact complete supported originals and their roles. |
| History | Admitted conversation/subject history relevant to the task. |
| Permissions | Actual granted access and any exact approval still required. |
| Pending operation | Original nonce/session/source, never a silently replaced request. |
| Private secrets | Keep in trusted custody; do not paste as ordinary context. |

## Retain a result another environment can inspect

Keep the enclosing source, operation references and supported history after a task completes. A model summary is useful but cannot become the Original or erase an unresolved action. Compatible environments admit carried identity/history using their own supported contracts and grants. Hosted compute, storage and tool availability retain their actual limits even when the identity protocol supports many distinct agents.

- [Continuity and subject generations](https://docs.bankeiz.com/agent-continuity/)
- [Exact integration boundaries](https://docs.bankeiz.com/integrations/)

## Implementation sources

- [Exact AI Skills release pin](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/package.json)
- [Coordinated release checks](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/scripts/receiz-release.mjs)
- [Skill/documentation references](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/docs/public-sdk-schemas.md)
- [Matching source archive](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/native/README.md)

## Related documentation

- [Your agent, identity and connected context](https://docs.bankeiz.com/agents/)
- [Give agents identity across environments](https://docs.bankeiz.com/agent-continuity/)
- [Expose supported operations through MCP](https://docs.bankeiz.com/mcp/)
- [Start with the Receiz SDK](https://docs.bankeiz.com/sdk/)