Developers · Agents

Give agents the canonical context before they write code.

Agent-built integrations fail when the agent guesses contract addresses, payment-method rules, webhook signatures, or delegation behavior. The fix is to feed the agent a compact, canonical context pack first.

Open the agent skillRead llms-full.txt
01

Agent context pack

ResourceUse
skills/usdctofiat.mdDrop-in skill with SDK rules, platforms, OTC, webhooks, and examples
llms.txtShort machine-readable product and route summary
llms-full.txtFull machine reference with contracts, pricing, guides, and answers
OFFRAMP_DEVELOPER_RESOURCESTyped SDK export with links, delegation config, and playbooks
Starters repoWorking Next.js, Vite, Telegram-bot, script, and webhook examples
02

Prompt sequence

  1. 1Tell the agent to read the skill and llms.txt before planning.
  2. 2Pick a profile: app, bot, agent, private-otc, webhooks, or peerlytics.
  3. 3Ask it to scaffold from create-offramp-app or the starters repo when possible.
  4. 4Require it to verify every created deposit delegates to the Delegate rate manager.
  5. 5If it adds webhooks, require raw-body HMAC verification before JSON parsing.
03

What agents must not invent

  • Do not invent a sandbox; the SDK targets Base mainnet.
  • Do not remove delegation; the SDK's deposit path must delegate to the Delegate vault.
  • Do not treat fiat as SDK-controlled. Fiat moves directly between buyer and seller.
  • Do not verify webhook signatures from parsed JSON. Verify timestamp.rawBody first.

Keep exploring

Offramp SDKWebhooksCash out by use caseDeveloper hub

Common questions

What should an agent read first?

Start with https://usdctofiat.xyz/skills/usdctofiat.md, then llms.txt or llms-full.txt depending on how much context the model can hold.

Can agents discover this from the package?

Yes. Import OFFRAMP_DEVELOPER_RESOURCES or call getOfframpDeveloperResources(profile) from @usdctofiat/offramp.