quickstart · ~10 min

From staging URL to first confirmed exploit.

The path of least resistance. Designed to land in your inbox before you finish your coffee.

Prerequisites

A staging environment we can reach over the public internet, or via a wireguard config you provide.
An OpenAPI 3.x spec or a recorded .har file describing the surface. Either works; both is better.
2–4 synthetic user accounts with credentials we can capture. Do not use real customer accounts.
The email you used to request access approved by us (usually same day).

why staging, not prod

Brink is configured against staging by default — we don't ship destructive verbs, but autonomous attackers are still autonomous. Prod targets are unlocked per project once you've shipped your first 14 days of findings.

1 · Point us at staging

From the dashboard, hit /projects/new. Give the project a slug, paste your base URL, and pick the environment. Slugs are lowercase, 1–48 chars; we use them in URLs and CI hooks.

brink cli# same as the dashboard form, scriptable
brink project create \
  --slug acme-api \
  --name "acme api" \
  --base-url https://api.staging.acme.dev \
  --env staging

2 · Set scope

Scope is the contract between you and the agent fleet. Anything not in scope, we don't touch. The defaults are sensible (no write verbs, 60 req/min, no admin paths), but you'll almost always tighten the path list and bump the rate.

scope.yamlbase_url: https://api.staging.acme.dev
allowed_paths:
  - /v1/users/*
  - /v1/orgs/*
  - /v1/billing/*
blocked_paths:
  - /v1/internal/*
  - /v1/admin/*
blocked_methods: [DELETE, PUT]
max_requests_per_minute: 240

scope is enforced at the proxy

Even if an agent reasons its way into a blocked path, the scope proxy refuses to forward the request. The agent gets a 403 from us, never your origin.

Synthetic accounts

Upload at least two accounts so we can probe for IDOR / authz crossings. We capture each account's bearer token (or session cookie) once and replay it from inside the proxy — your origin only ever sees its own credentials.

brink clibrink scope account add \
  --project acme-api \
  --label "tenant-a" \
  --token "$(cat ./fixtures/tenant-a.bearer)"

3 · The first 24 hours

Once scope and accounts are in, the recon agent wakes up immediately. The first thing you'll see is hypotheses fading into the /ops feed — guesses about where the interesting attack surface lives. Confirmed exploits arrive over the next 14 minutes to 24 hours depending on your surface area.

~14m: the first hypothesis confirms. Severity is whatever the validator measures, not the agent's guess.
~1h: coverage report shows what we tried (and skipped).
~24h: 80% of the "easy" surface is mapped. Agents settle into a tighter probing loop on the harder paths.
day 3+: agents start chaining primitives — an info leak feeds an IDOR feeds a takeover.

What to do next

When your first confirm hits, the inbox row will have a reproducible curl, a captured response, a suggested patch, and (once you've connected GitHub) an open PR against the offending repo. Read the concepts page next if you want to know how any of that is wired.