Skip to main content
Each Paxos SDK ships with CLAUDE.md and AGENTS.md files that provide structured context for AI coding assistants. These files help LLMs generate correct SDK code by describing available methods, patterns, and common pitfalls.

​
How It Works

When you point an AI assistant at a Paxos SDK repository, the CLAUDE.md file gives it:
  • A summary of all available services and methods
  • Correct patterns for authentication, error handling, and pagination
  • Common workflows (e.g., “create identity, approve it, fund the profile, make a transfer”)
  • Warnings about common mistakes (e.g., using floats for monetary amounts)

​
Using Claude with the SDK

If you’re using Claude Code, the CLAUDE.md file is read automatically when you work in the SDK repository. Example prompts that produce good results:
  • “Create a crypto withdrawal with idempotency from profile X to address Y”
  • “Set up a webhook handler that processes identity approval events”
  • “Write an integration test that deposits USD and lists transfers”
  • “Implement pagination to export all transfers to a CSV file”

​
LLM-Friendly Documentation

All SDK methods include natural language docstrings with business context, not just type signatures. This helps LLMs understand when to use a method, not just how. For example, asking an LLM to “implement a crypto withdrawal with idempotency” produces code that correctly uses CreateCryptoWithdrawal with an idempotency key — because the docstring explains the business context and the idempotency pattern.

​
AGENTS.md Reference

Each SDK includes an AGENTS.md file with task-specific instructions for common workflows:
  • Setting up a new integration from scratch
  • Adding webhook handling to an existing application
  • Migrating from direct API calls to the SDK
  • Writing integration tests against the Sandbox

​
Best Practices

  • Use environment variables for credentials — Never paste PAXOS_CLIENT_ID or PAXOS_CLIENT_SECRET inline in prompts
  • Test generated code in Sandbox — Always verify generated code works in Sandbox before deploying to production
  • Use string types for amounts — LLMs sometimes suggest floating-point types for monetary amounts. Paxos SDKs use string-based decimal types to avoid precision errors
  • Review error handling — Verify that generated code handles errors appropriately for your use case

​
Documentation Resources

Paxos provides machine-readable documentation for AI tools:
  • /llms.txt — Structured index of all documentation pages, following the llmstxt.org convention
  • /llms-full.txt — Complete documentation compiled into a single file for full-context AI prompts
  • OpenAPI specification — Complete REST API schema for precise implementation details
For more about using AI tools with Paxos APIs generally, see the AI-Assisted Development guide.