email-deliverability — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited email-deliverability (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
A bulleted imperative like {match} tells the agent to never reveal, disclose, or mention something to the user. Used adversarially it can instruct the agent to hide its tool calls or lie about what it did — stripping the transparency a user relies on to trust the agent.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
When configuring email sending infrastructure, troubleshooting delivery issues, warming up new sending domains/IPs, or architecting transactional vs marketing email separation. Use for any task that affects whether emails reach the inbox.
Core principle: Reputation is everything — email deliverability is a long game. One bad send can destroy months of reputation building. Protect sender reputation like you protect production uptime.
; Authorize sending IPs/services
v=spf1 include:_spf.google.com include:sendgrid.net include:amazonses.com -allRules:
-all (hard fail) not ~all (soft fail) for production domainsinclude sparingly+all (allows anyone to send as you) ; Public key for signature verification
selector1._domainkey.example.com IN TXT "v=DKIM1; k=rsa; p=[public_key]"Rules:
; Tell receivers what to do with failures
_dmarc.example.com IN TXT "v=DMARC1; p=reject; rua=mailto:[email protected]; ruf=mailto:[email protected]; pct=100"Deployment progression:
Week 1-2: p=none (monitor only, collect reports)
Week 3-4: p=quarantine; pct=10 (quarantine 10% of failures)
Week 5-6: p=quarantine; pct=50
Week 7-8: p=quarantine; pct=100
Week 9+: p=reject (full enforcement — unauthenticated mail rejected)Rules:
Day 1-3: 50 emails/day (to most engaged recipients only)
Day 4-7: 100 emails/day
Week 2: 200-500/day
Week 3: 500-1,000/day
Week 4: 1,000-5,000/day
Week 5: 5,000-10,000/day
Week 6+: Increase 2x per week until target volumeRules:
| Metric | Healthy | Warning | Critical |
|----------------------|------------|------------|------------|
| Bounce rate | <1% | 1-2% | >2% |
| Complaint rate | <0.05% | 0.05-0.1% | >0.1% |
| Open rate | >20% | 10-20% | <10% |
| Spam trap hits | 0 | 1-2/month | >2/month |
| Blacklist presence | None | 1 minor | Major list |Actions:
Hard bounce (permanent failure):
- Invalid address, domain doesn't exist, mailbox doesn't exist
- Action: Remove from list IMMEDIATELY (first occurrence)
- Never retry a hard bounce
Soft bounce (temporary failure):
- Mailbox full, server temporarily unavailable, message too large
- Action: Retry up to 3 times over 72 hours
- After 3 soft bounces on same address: treat as hard bounce and suppress
Complaint (user clicked "spam"):
- Action: Suppress IMMEDIATELY, never email again
- Process FBL (Feedback Loop) reports within 1 hour
- If complaint rate rises: review recent sends for consent issues - Remove hard bounces: immediately
- Suppress complaints: immediately
- Remove unengaged: no open/click in 90 days → sunset sequence → remove
- Validate on signup: real-time email validation API (catch typos, disposable domains)
- Re-validate periodically: quarterly bulk validation of full list
- Double opt-in: recommended for all marketing (required in some jurisdictions) Transactional email (receipts, password resets, 2FA):
- Dedicated IP/subdomain: mail.example.com
- Priority: immediate delivery (no batching)
- Volume: consistent, predictable
- Reputation: protected (never mixed with marketing)
Marketing email (newsletters, promotions, re-engagement):
- Dedicated IP/subdomain: news.example.com
- Priority: send-time optimized (batch by timezone/engagement)
- Volume: variable, seasonal spikes
- Reputation: more volatile (isolated from transactional)Rules:
Before marking a task done when this skill was active:
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.