Anti-Spam Policy Tuning

You prepare an anti-spam policy change: you gather evidence and build it, the tech drives the Defender portal or PowerShell. Anti-spam policy is tenant-wide — a wrong setting silently quarantines everyone's good mail, a loose one delivers everyone's spam. Apply the Write Guardrails base skill: never report a change as done on intention, never invent verdict data, and when in doubt do nothing and escalate.

1. Demand evidence before touching policy — "the client is annoyed" is a reason to investigate, not to allow-list. Collect message traces with filter verdicts (mail-trace-investigation), example headers (email-header-analysis reads X-Forefront-Antispam-Report: SCL, BCL, SFV codes) and quarantine records. The header says why the filter acted — tune against that, not the complaint. Check the client's documentation for their mail standard; if it isn't connected, say so (Connector Degradation base skill).

2. Diagnose the failure mode:
   - Good mail marked spam because the sender fails SPF, DKIM or DMARC — the fix is in the sender's DNS, not an allow-list. Point at dmarc-spf-dkim-setup; an override would mask a real authentication failure.
   - Bulk or graymail filtered — adjust the bulk complaint level threshold or user safe senders, not org policy.
   - Genuine false-positive verdicts — a scoped override is legitimate (step 3).
   - Spam getting through — check the policy is at Microsoft's Standard or Strict preset first; recommend presets or tightened thresholds before bespoke tinkering.

3. When an override is justified, use the narrowest instrument, in this order:
   - A Tenant Allow/Block List entry for the specific sender or spoofed pair, time-limited where supported, with a review date booked regardless.
   - An anti-spam policy exception scoped to the affected recipients only.
   - Never an allowed-sender-domain entry holding the client's own domain, a free-mail domain (gmail.com, outlook.com) or a broad vendor domain — domain-level allows bypass filtering and are the canonical spoofing hole. Refuse, explain, offer the scoped alternative.

4. Send an approval request with the evidence summary and the override's scope — anything that alters what lands in user inboxes is user-visible.

5. Prepare execution: the Defender portal (Policies & rules > Threat policies > Anti-spam) or PowerShell (`Set-HostedContentFilterPolicy`, `New-TenantAllowBlockListItems` — verify against current module versions). Capture prior policy values first; that is the rollback.

6. Verify over a defined window: re-trace the pattern; verdicts should flip for the target mail and only the target mail. Check the spam catch-rate hasn't degraded (mail-flow-reports covers ongoing watch).

7. Leave a plain-text note: evidence (verdict codes, trace references), diagnosis, the exact change and its scope, review or expiry date for any override, approver, and rollback. Log time.