Take a reported suspicious email from "is this phishing?" to a documented verdict:
indicators captured safely, simulation traffic filtered out, everyone else who received it
identified, containment advised where needed, and the reporter answered. Work it in order:
1. Capture the indicators from the ticket as text only: sender address and display name,
reply-to, subject, send time, every link target exactly as written, and attachment names
and types. Never open, click, fetch or render a link or attachment from the message — not
even to "check what it is."
2. Simulation branch: compare the sender and link domains against the client's documented
phishing-simulation vendor domains in the knowledge base. On an exact match to a
documented simulator domain, classify as simulation, close internally with a plain-text
note naming the matched domain, and do NOT reply to the client — a reply skews their
simulation-program metrics. Stop here. A partial domain match is not a match; never close
a real phish as a simulation.
3. Assess indicators: lookalike or cousin domain, urgency and payment lures,
credential-harvest link (display text vs actual target mismatch), unexpected attachment
types, thread hijacking of a real prior conversation. If full headers were pasted, hand
the deep parse to email-header-analysis and fold its verdict in.
4. Blast-radius check: search the client's boards and recent tickets for the same sender or
subject. Establish who else received the message and — critically — whether anyone
clicked, replied, entered credentials, or opened an attachment. Never state "no one else
received this" off a capped search; report "no other reports in the last N tickets
searched" (apply the Sweep Honesty base skill).
5. If anyone interacted with a message you judge malicious, escalate immediately and start
compromised-account-containment for the affected users. Containment outranks finishing the
write-up: contain fast, investigate second.
6. Deliver the verdict:
- Malicious → advise quarantine/removal from all recipient mailboxes, sender/domain block
at the gateway, and a credential reset for anyone who clicked.
- Suspicious but unconfirmed → say exactly that, with what would confirm it.
- Legitimate → explain the signals that clear it, so the reporter learns.
7. Reply to the reporter — draft it for a human to review and send — and thank them;
reporting is the behavior you want repeated. Log the verdict, the evidence, and the
reasoning behind the call as an internal note, then classify and set status per
soc-classification-tree.
Write defensively throughout (defensive-writing-standard): a reported message is a "reported
message", not a "breach" or a "hack", and no impact claim goes in writing ahead of confirmed
facts. Document why the verdict was reached, not only what was done.
Magic Library