@didwork/sdk

Your software said it worked. Did it?

DidWork independently verifies that the outcomes software claims to have produced are actually true. Send a claim; DidWork gathers evidence from authoritative systems and returns a verdict: verified, failed, or unknown.

import { did } from "@didwork/sdk";

const result = await did.verify({
  type: "stripe.refund",
  expected: { payment: "pi_123", amount: 4999 },
});

if (result.status === "verified") proceed();

Configure with DIDWORK_API_KEY (get one at didwork.sh), or explicitly via createClient({ apiKey }). Zero dependencies.

Capability Trust

Declare what an agent tool is allowed to affect. did.tool() is local and synchronous: it attaches the declaration and rejects effects outside the DidWork taxonomy, so a typo cannot become an under-declaration.

export const refundCustomer = did.tool({
  name: "refund_customer",
  description: "Refund an existing customer charge.",
  effects: ["financial.refund", "communication.external"],
  constraints: { maxAmount: 500 },
});

npx @didwork/inspect inspect compares declarations like this against the implementation and reports a DECLARATION_MISMATCH when the code exceeds them. Pass the tools that did a piece of work to did.verify({ tools }) and the receipt records what they could affect, kept apart from the evidence and never an input to the verdict:

await did.verify({
  type: "stripe.refund",
  expected: { payment: "pi_123", amount: 4999 },
  tools: [refundCustomer],
});

The taxonomy (CAPABILITIES), risk table (CAPABILITY_RISK), and profile types are exported for tooling.

Docs: https://didwork.sh/docs