spam-filter-advisorlisted

Analyze a folder of saved spam emails (.eml files) and recommend mail filter rules that catch them without eating legitimate mail. Use whenever someone has collected spam samples and wants help building filters, says their inbox is flooded with junk and they want rules to stop it, asks what patterns their spam shares, wants to know which keywords or sending IPs are safe to block, or wants help setting up rules in Apple Mail, Gmail, Outlook, or Thunderbird. Also when someone exports spam to a folder and asks "what can I do about this" without naming filters. Equally for tuning rules already in place — spam still slipping through, a filter catching real mail by mistake, asking why a message got flagged, or returning with a folder of misses and false alarms. Produces recommendations ranked by false-alarm risk with exact copy-paste strings, warns when the folder is contaminated with real mail, and can test existing rules to find which one misfired.
idea2go2go/spam-filter-advisor-skill · ★ 2 · Data & Documents · score 70

Install: claude install-skill idea2go2go/spam-filter-advisor-skill

# Spam Filter Advisor Someone has a folder of spam and wants filter rules. The job is to find the patterns that are genuinely unique to the spam, rank them honestly by how likely they are to catch real mail by mistake, and hand over rules the person can actually install. The reason this needs care rather than pattern-matching enthusiasm: a filter that misses spam is an annoyance, but a filter that quietly diverts a job offer or a medical result is a real harm the person may not discover for weeks. Those two failure modes are not symmetric, and the recommendations should not treat them as if they were. When in doubt, recommend the narrower rule. ## Workflow ### 1. Find the folder and confirm what is in it Ask where the .eml files are if it is not obvious. Confirm the person understands these should be **spam only** — messages they are confident are junk. If they say they dragged in "everything from the last week" or seem unsure, flag now that mixed-in real mail will produce filter rules that catch real mail. **How many samples.** Recommend **30 to 50 recent examples**. That is the range where frequency counts, IP clustering, and misspelling patterns all become dependable enough to build rules on. Below 30, say so before running and name what gets weaker: frequency counts and IP clustering are the first things to become unreliable, while character-substitution findings (Tier 1) stay trustworthy at any size, because a deliberate lookalike misspelling is self-evidently del