Brand scoping

Outputs live under ~/notes/Klaviyo/<brand>/<subfolder>/, where <brand> is resolved at run time from the connected Klaviyo account's organizationName attribute (via mcp__claude_ai_Klaviyo__klaviyo_get_account_details). For ExampleBrand, this resolves to ExampleBrand. If multiple brands are connected, ask Will which brand the current run is for.

/klaviyo-deliverability

Email that lands in spam is email that doesn't exist. This skill runs the monthly health check that catches deliverability problems BEFORE they kill list quality — bounce/complaint trends, auth records (DKIM/SPF/DMARC), engagement decay, sender domain reputation.

Flow

1. Pull account-level numbers — mcp__claude_ai_Klaviyo__klaviyo_get_account_details for sender domains + authentication status. Read the Klaviyo dashboard's deliverability tab via the account details if exposed.
2. Compute bounce rate (last 30 days) — mcp__claude_ai_Klaviyo__klaviyo_query_metric_aggregates for Bounced Email / Email Delivered. Threshold: hard bounce rate must be <2%. Above = list hygiene problem.
3. Compute complaint rate (last 30 days) — same approach with Marked Email as Spam. Threshold: <0.1%. Above = list quality OR content problem (over-mailing, irrelevant content, surprise senders).
4. Compute unsubscribe rate (last 30 days) — Unsubscribed events. Threshold: <0.5%. Above = cadence too high OR content irrelevance.
5. Pull engagement decay — opens last 30d vs 60d vs 90d. Sharp decline = sender reputation eroding OR fatigued list.
6. DKIM / SPF / DMARC check — for the brand's sending domain. Run dig against _dmarc.<domain>, <selector>._domainkey.<domain>, <domain> for SPF. Surface missing or weak records (DMARC p=none = no enforcement).
7. Sender reputation lookup — for each sending IP/domain, check public reputation sources if accessible (talos, senderscore, microsoft snds). Note: most reputation APIs are gated; flag manual-check items if automation isn't possible.
8. Compute composite health score — 0-10. Components: bounce score (3 pts), complaint score (3 pts), unsubscribe score (2 pts), auth score (DKIM+SPF+DMARC, 2 pts).
9. Recommend fixes — ranked by impact. For each: what to do, where in Klaviyo to do it, expected lift.
10. Save — ~/notes/Klaviyo/<brand>/deliverability/YYYY-MM-DD.md. Append trends from prior months.
11. Print summary — health score, 3 top fixes, link to file.

Health thresholds (the red zones)

Common fixes (ranked by impact)

1. Sunset disengaged subscribers — if open rate is dropping or complaint rate rising, suppress profiles with no engagement in 90+ days. Often a 30-50% engagement-rate lift overnight (and complaint rate craters).
2. Upgrade DMARC from p=none to p=quarantine — gradual rollout (10% → 50% → 100%). Most brands sit on p=none indefinitely. Quarantine + the right alignment dramatically improves inbox placement.
3. Add DKIM for the sending domain (not Klaviyo's default) — Klaviyo lets you sign with their domain by default, but inbox placement is better when DKIM signs with the brand's domain.
4. Reduce send cadence to fatigued segments — if unsubscribe rate is climbing, throttle sends to the lowest-engagement tier.
5. Re-engage cold subscribers BEFORE suppressing — /klaviyo-flow win-back first, then sunset. Some come back.
6. Switch to a dedicated IP if sending >100k emails/month — shared IPs commingle reputation across all senders. Dedicated isolates.

Output structure

# Klaviyo Deliverability — [brand] — YYYY-MM-DD

## Composite health score: N/10 (trend: +/- from last month)

## Metrics (last 30 days)

| Metric | Value | Status | Trend |
|---|---|---|---|
| Hard bounce | X% | green/yellow/red | +/- |
| Soft bounce | X% | ... | ... |
| Complaint | X% | ... | ... |
| Unsubscribe | X% | ... | ... |
| Open (engaged) | X% | ... | ... |

## Authentication

- DKIM: passing/failing for [domain]
- SPF: ...
- DMARC: p=... rua=...

## Top 3 fixes ranked by impact

### 1. [Fix]
**Issue:** ...
**Action:** ...
**Expected lift:** ...

### 2. ...
### 3. ...

## Trend from prior months
- ...

What this skill does NOT do

• Does not auto-suppress profiles or change account settings — surfaces recommendations, Will applies.
• Does not check inbox placement vs spam folder via seedlist (that's a paid service like Litmus/GlockApps).
• Does not do real-time monitoring — monthly cadence by design.

Failure modes

• MCP returns no deliverability stats → fall back to query_metric_aggregates for bounces/complaints/unsubs from event metrics; surface what's unknowable.
• No sending domain in account details → ask Will to confirm the domain so DMARC/DKIM checks can run.
• Composite score didn't change month-over-month → say so, don't manufacture a trend.

Cadence

Run monthly, first business day. Can be scheduled via /schedule or cron.