Case file
Rebuilding a phishing kit in a lab, safely
Defenders who have never driven the attacker’s tooling tend to write detection rules about the wrong things. So this case file does what the Sentinel’s rules require of every offensive writeup: it rebuilds the technique entirely inside a lab, on infrastructure the Sentinel owns, with no real sending and no real targets — and then turns the experiment into detection engineering a defender can reuse.
The Sentinel’s Rules, Rule 1 and Rule 5, apply in full: this build runs on an isolated network with synthetic users and a mail path that never touches the public internet. Nothing here is a phishing kit ready to aim at people; it is a receiver-side telescope. See The Sentinel’s Rules and Disclosures.
The tool: GoPhish, honestly introduced
GoPhish is an open-source phishing-simulation framework — MIT-licensed, maintained in public, and used legitimately by security teams to run authorized awareness exercises. That last clause is the load-bearing one: the framework is neutral, the authorization is not optional, and this file’s entire build operates in a lab where the only “targets” are synthetic mailboxes the Sentinel created for the purpose.
MITRE ATT&CK classifies the real-world behavior this tooling automates as T1566, Phishing, and defenders benefit from knowing its shape: a sender, a lure, a landing page, and a credential capture. The lab reproduces all four so that the receiving side can be studied honestly.
The build: isolated by construction
The environment is boring on purpose, and its boredom is the safety mechanism:
- Network: an isolated segment with no route to the internet. The mail server and the GoPhish instance live inside it; nothing leaves.
- Sending: GoPhish’s built-in sending profile points at a lab-internal SMTP relay. There is no external relay, no rented domain that has ever sent mail, and no warm infrastructure — deliberately, because the point is to study what such mail looks like on arrival, not to make it arrive anywhere real.
- Targets: synthetic mailboxes (
alice@,bob@) on a lab-only mail domain. No real addresses exist in the environment at any point. - The lure: a deliberately generic “quarterly document request” template — mild enough that its purpose is education, not entrapment. The Sentinel’s rules draw a hard line here: an awareness exercise measures process, it does not set people up to fail.
What the receiving side actually sees
Here is the transferable part. When the lab relay delivers the simulated message to the synthetic mailbox, the interesting content is not the lure — it is the headers, and the three-letter acronyms that decide whether the message would ever have reached a real inbox:
- SPF (RFC 7208) asks: is the sending server authorized by the domain in the envelope? The lab fails this immediately — the lab’s IP appears in no SPF record, because no legitimate one exists.
- DKIM (RFC 6376) asks: is the message cryptographically signed by the domain it claims, and does the signature cover what arrived? The lab message is unsigned, and unsigned is what the receivers’ logs record.
- DMARC (RFC 7489)
asks: did SPF or DKIM pass and align with the visible From domain?
This is the tripwire that fires for the lab build, and in production it
is the tripwire that fires for most commodity phishing — which is why
sender domains that enforce a DMARC policy of
p=rejectare so much harder to impersonate convincingly.
The practical exercise the Sentinel runs with this build: deliver the
message, open it in the synthetic mailbox, and walk the full header chain
— Authentication-Results lines, alignment failures, the gap between the
envelope domain and the display name. Every defender should have read one
raw phishing header end to end; the lab provides one whose origin is
completely accounted for.
The metrics worth measuring
Organizations that run authorized simulations with tooling like GoPhish usually fixate on the click rate, and that fixation is the error. The number worth managing is the report rate — how many users forward the message to the security team or hit the report button — because a clicked lure that gets reported in minutes is a success, and an unclicked lure nobody reported is a quiet failure. Click rates measure fear; report rates measure process. CISA’s phishing guidance (2023) frames the defensive goal the same way: catch the attack at its first phase, when it is still an email and not yet an incident.1
No individual is ever named in a Sentinel-style exercise report. The output is one chart: reports per campaign, trending up.
The transferable sentence
The phishing kit is not the threat; the untested receiving path is. A defender who has stood up the tooling in a lab — seen the failed alignment in the headers, felt how small the tells are — writes mail gateway rules, DMARC policies, and awareness programs that aim at what attackers actually ship. And every part of that education fits inside a network with no route to the outside world.
All techniques in this file were demonstrated in the Sentinel’s isolated lab on synthetic targets, per The Sentinel’s Rules. The security posture of this site’s own mail and headers is documented in its headers case file.
Sources & attributions
- GoPhish — open-source phishing simulation framework (MIT license) — getgophish.com.
- MITRE ATT&CK, T1566: Phishing — attack.mitre.org/techniques/T1566.
- CISA, Phishing Guidance: Stopping the Attack Cycle at Phase One (Oct. 2023) — cisa.gov.
- RFC 7208 (SPF), RFC 6376 (DKIM), RFC 7489 (DMARC) — IETF — datatracker.ietf.org.
Footnotes
-
CISA’s guidance organizes defenses around the first phase of the phishing attack chain — provisioning, sending infrastructure, and the message itself — before credential capture becomes possible. ↩