security-testinglisted
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# Security Testing
Scope this honestly. What a QA engineer with an agent can do is verify
authorisation boundaries, check for the well-known classes of mistake, and find
what a scanner finds. What it cannot do is a penetration test. Say so in the
report — a document that implies more assurance than it delivers is worse than no
document.
**Never test a system you do not have written permission to test.** Never run
active scanning against production. Never attempt to access another real user's
data — use test accounts you control on both sides of every authorisation check.
## Start with authorisation, because that is where the bugs are
The OWASP API Security Top 10 (2023 edition, still the current list) puts three
authorisation failures in the top three, and the first of them is the single most
common serious API bug. Test them in this order; it matches where findings
actually are.
**BOLA — broken object level authorisation (API1).** Take an object ID belonging
to user A, request it as user B. This one line is the highest-value security test
in existence for a typical API.
```
As user B: GET /api/v1/orders/{A's order id}
Expected: 404 or 403. A 200 is a finding, and usually a critical one.
```
Test it on every endpoint that takes an ID — read, update **and** delete. Teams
that fixed it on GET routinely left DELETE open. Try the ID in the path, in a
query parameter, in the body, and in a nested field.
**Broken authentication (API2).** Expired token accepted, token si