Proof Docs

Repositories

Connect one codebase, its browser target, and the narrow network and evidence policy Proof needs to discover business coverage.

Connect a repository

POST /api/repositories

json
{
  "provider": "github",
  "providerInstallationId": "123456",
  "owner": "acme",
  "name": "storefront",
  "defaultBranch": "main",
  "cloneUrl": "https://github.com/acme/storefront.git",
  "baseUrl": "https://storefront.example.com",
  "networkAllowlist": ["api.stripe.com", "evidence.acme.test"],
  "testValueSecretRefs": {
    "orderProbeUrl": "PROOF_ACME_ORDER_PROBE_URL",
    "orderProbeToken": "PROOF_ACME_ORDER_PROBE_TOKEN"
  },
  "evidenceProbeRefs": {
    "order-created": {
      "urlValueRef": "orderProbeUrl",
      "tokenValueRef": "orderProbeToken"
    }
  }
}

The API key determines the organization. GitHub ingestion is allowed only after an administrator connects that installation to the same organization in the dashboard. Permanently excluded owners are rejected before persistence or billable ingestion.

Configuration fields

FieldDescription
owner / nameRepository coordinates used by GitHub webhooks.
cloneUrlHTTPS clone URL. Private access uses the GitHub App installation.
baseUrlDefault browser target for discovery and replay.
networkAllowlistAdditional hostnames explicitly available to browser runs.
testValueSecretRefsMap of semantic test values to runtime environment-secret names.
evidenceProbeRefsMap success criteria to approved read-only HTTPS evidence probes.
revisionOptional commit SHA for the initial repository model.

Configure database-effect evidence

Required database effects use a customer-controlled read-only HTTPS probe. The probe returns a bounded response such as {"passed":true,"evidence":"Order NS-2048 exists"}. Proof gives only this result to the verifier and compiles the same check into the deterministic test.

Response and ingestion

A valid request returns 202 Accepted with the key's organization, repository, ingestion run, and Workflow run identifiers. Repository onboarding then indexes the requested revision, discovers flows and edge cases, creates coverage links, and starts bounded child catalog builds.

This route requires both live execution and AI execution because initial repository discovery creates a Sandbox and invokes the planner.