salesloft-sdk-patterns
Analyze and design a typed Salesloft REST adapter with tenant-bound authentication, endpoint-specific schemas, pagination, errors, and rate metadata. Use when creating or refactoring an integration client. Trigger with "Salesloft SDK patterns", "Salesloft client wrapper", or "typed Salesloft API".
Allowed Tools
Provided by Plugin
salesloft-pack
Production Salesloft operator workflows for authentication, cadences, synchronization, rate control, webhooks, security, deployment, and incident evidence
Installation
This skill is included in the salesloft-pack plugin:
/plugin install salesloft-pack@claude-code-plugins-plus
Click to copy
Instructions
Salesloft Typed REST Adapter
Overview
This skill builds a narrow client around the endpoints an application actually uses. It avoids invented universal schemas, implicit tenant state, and unbounded recursive retries.
Prerequisites
- A repository with an identified HTTP boundary
- Endpoint, scope, request-format, and response-format inventory
- A tenant-to-credential resolver
- Fixture coverage for success and failure envelopes
Tool Discipline
Use Read, Glob, and Grep to inspect package versions, call sites, and generated types. Use WebFetch only for official endpoint contracts. Use Write or Edit after the adapter boundary is confirmed.
Current Contract
- Use
https://api.salesloft.com/v2and documented resource paths; do not force.jsononto every path. - Preserve
data,metadata,error, and field-keyederrorsas separate typed shapes. - Pagination and filters are endpoint-specific; page numbers begin at 1 and deep pages consume more rate cost.
- Capture
x-ratelimit-endpoint-costandx-ratelimit-remaining-minutefrom every response that provides them. - Use the request content type documented by the endpoint rather than assuming all writes are JSON.
Authentication
Resolve one Bearer credential from the explicit tenant context for every call. Never store a mutable global token or retry with another tenant's credential.
Instructions
- Inventory exact methods, paths, scopes, query parameters, bodies, and consumed response fields.
- Define typed success, list metadata, singular error, validation error, and rate-state models.
- Centralize timeout, Bearer injection, content negotiation, and redacted error capture.
- Implement an iterative page iterator that stops on
metadata.paging.next_pageor an empty page. - Retry only safe operations or explicitly idempotent application workflows with a bounded attempt budget.
- Add contract fixtures before migrating callers behind the adapter.
Approval Boundaries
Do not synthesize request fields from examples, automatically fall back to v1, or retry CRM writes without an application idempotency rule and operator approval.
Output
Return endpoint inventory, type boundaries, auth resolver, pagination behavior, retry policy, fixture coverage, and remaining direct calls.
Error Handling
| Condition | Response |
|---|---|
| 401 | Refresh or reacquire through the configured auth flow once. |
| 403/404 | Preserve singular error and verify scope, visibility, and path. |
| 422 | Preserve field-keyed errors; do not flatten away field context. |
| 429/5xx | Respect current headers and bounded backoff; surface exhaustion. |
Examples
The example below shows the minimum redacted evidence expected from a successful invocation of this operator workflow.
tenant=team-42; endpoint=GET /v2/people; pages=2; rate-cost=2; retries=0