Nurture Health

design-nurture-sequence drafts it, wire-nurture-sequence ships it — and then nobody looks again. This skill is the part that was missing: it tells the operator whether the sequence is actually working, which step is bleeding people, and whether the machine's un-enrol side is firing so booked/dead leads stop getting drip emails.

It pairs with the re-engagement machine in knowledge/automation-recipes.md (R19/R20) — that section explains what a healthy multi-channel queue looks like.

Step 1 — Fetch the health digest

python ~/.claude/bos-run.py nurture-health

One run: lists every auto queue, pulls each one's steps + enrolment funnel + per-step drop-off, and (where the email-log endpoint exposes automation_id) open/click rates per step. Scope to one queue with --queue <id>.

This fetcher is best-effort by design — queue / enrolment / email-log endpoints vary by workspace and API version. Anything it can't reach lands in warnings and _sources, and it still returns everything else. Read those two fields and tell the operator plainly what's measured vs estimated.

Fallback if the script can't run at all (auth/network): drive it by hand — list_auto_queues → get_auto_queue (steps) → get_auto_queue_board (per-step buckets) → list_auto_queue_enrollments (status mix) → list_email_logs (engagement). That's many calls; prefer the script.

Step 2 — Present, worst leak first

Lead with the single biggest leak across all queues, then go queue by queue. Never dump raw step arrays — translate them into a funnel the operator reads in ten seconds.

📉 BIGGEST LEAK: "Reawakening Sequence" loses 48 people at Day 7 (step 2).
   174 enrolled → 126 made it to Day 7. That's where to look first.

♻️ Reawakening Sequence  (active)
   Funnel: 174 enrolled → 120 active · 30 completed (17%) · 24 cancelled
   Steps:
     Day 0   174 reached   open 62%  click 18%
     Day 7   126 reached   open 41%  click 7%   ← 48 dropped here
     Day 14   118 reached   open 38%  click 6%
     ...
   ✅ Un-enrol firing: 24 cancelled means booked/dead leads are leaving the drip.

📨 Onboarding Sequence  (active)
   Funnel: 90 enrolled → 60 active · 22 completed (24%) · 8 cancelled
   ...

⚠️ Couldn't measure: open/click rates (email logs don't link to automations in
   this workspace) — funnel + drop-off are exact, engagement is unavailable.

How to read each signal

The verdict layer (what the numbers mean for the business)

Read each queue against the doctrine lens that fits its job: an onboarding/enrolment queue against the activation lens (business-method.md §11.3 — the % reaching the first win early IS next quarter's churn, directional); a retention sequence against the ceiling equation (§11.1 — churn caps the business's size regardless of marketing); a lead/reawakening sequence against the engaged-lead math (§10.3 — fix the WORST stage, not the top). The funnel says where the leak is; the lens says what it costs.

Step 3 — Point at the fix, don't auto-fix

This is a read/diagnose skill. For the fixes it surfaces, hand off:

• Leaky/weak copy → /design-nurture-sequence (rewrite), then /wire-nurture-sequence.
• Style/consistency problems → /lint-nurture-sequence.
• Un-enrol not firing / a step never sends → /why-didnt-it-fire, then /automate-this.

Don't edit queue steps or automations from this skill.

What to never do

• ❌ Don't present queues as a flat list — lead with the biggest leak, then per queue.
• ❌ Don't report open/click rates as exact when _sources.engagement isn't ok — say "estimated" or "unavailable".
• ❌ Don't call a queue "broken" because completion is low — low completion can be a long sequence working as designed. Flag the leak step, not the headline rate.
• ❌ Don't write to any queue or automation here.

Output shape

Open with the single biggest leak (queue + step + how many lost). Then one compact funnel block per queue. Then any "couldn't measure" line from warnings/_sources. Close with the one step most worth fixing first and the skill to hand it to.