crewly 1.20.447 → 1.20.449

package/config/skills/agent/checkup-analyse/SKILL.md

@@ -0,0 +1,99 @@
+---
+name: Check-up Analysis
+description: The analysis step of the small-business check-up. Reads what ingest-exports wrote, links the same person across every file, follows inquiry → sign-up → payment → repeat and lists where it breaks (unanswered inquiries, sign-ups with no payment, payments from nobody on the roster, drop-outs, lapsed customers, overdue invoices, double entry, customers missing from the mailing list), each with an hours/week and $/month estimate whose formula is shown. Writes breaks.json and report.md.
+version: 1.0.0
+category: analysis
+skillType: claude-skill
+assignableRoles:
+  - "*"
+triggers:
+  - check-up analysis
+  - analyse check-up
+  - find the breaks
+  - where does the business leak
+  - checkup report
+  - top 3 breaks
+tags:
+  - check-up
+  - small-business
+  - analysis
+  - report
+execution:
+  type: script
+  script:
+    file: execute.sh
+    interpreter: bash
+    timeoutMs: 120000
+---
+
+# Check-up Analysis
+
+Step 3 of the small-business check-up (crewly#1062). After `ingest-exports`
+has turned the owner's CSV exports into one schema, this skill reads that
+folder and says where the chain **inquiry → sign-up → payment → repeat**
+breaks between the apps, and what each break costs.
+
+```bash
+bash execute.sh --in "$CREWLY_HOME/checkup/<customer>/ingest" --out "$CREWLY_HOME/checkup/<customer>/analysis"
+bash execute.sh --in …/ingest --as-of 2026-10-31 --assume hourlyValue=40 --assume lapsedDays=60
+bash execute.sh '{"in":"…/ingest","out":"…/analysis","asOf":"2026-10-31","assume":{"winBackRate":0.1}}'
+```
+
+`--in` is the `--out` folder of `ingest-exports` (or single `<file>.json`
+files from it). `--as-of` defaults to the last dated activity in the data;
+pass today's date when the exports are fresh so overdue and lapsed are
+measured against now.
+
+## The breaks
+
+| type | what it means |
+|---|---|
+| `inquiry_no_follow_through` | a form submission (Jotform) from someone with no sign-up and no payment at any time |
+| `signed_up_unpaid` | a sign-up (Jackrabbit, Mindbody, Calendly) with no payment from that person on or after it (7-day grace before) |
+| `paid_not_listed` | a payment from someone on no roster (only checked when a roster export is present) |
+| `dropped` | a sign-up marked dropped / cancelled |
+| `lapsed` | a customer with no activity for `lapsedDays` (90) before as-of |
+| `overdue_invoices` | an invoice with a balance whose due date is before as-of |
+| `double_entry` | the same person (by email, then phone) in two or more systems; notes a phone missing in one or differing names on contact rows |
+| `not_on_mailing_list` | a paying or signed-up customer the mailing list (Mailchimp) does not have (only when a list export is present) |
+
+People are linked by email first, then phone, then name when a row has
+neither. Hashed exports (`ingest-exports --hash-names`) link the same way,
+so the counts are identical and no name appears in the output.
+
+## Estimates
+
+Every break carries `estimate: { hoursPerWeek, dollarsPerMonth, formula,
+isEstimate: true }`. The formula is written out with the numbers used, e.g.
+`$180 in unpaid fees ÷ 1 months = $180/month; 1 × 10 min ÷ 3.3 weeks = 0.1
+h/week`. The inputs are assumptions the owner can change with
+`--assume key=value`:
+
+- `hourlyValue` 30 — what an hour of the owner's or staff's time is worth
+- `minutesPerManualEntry` 4, `minutesPerFollowUp` 10
+- `inquiryConversionRate` 0.3, `winBackRate` 0.2, `listReachRate` 0.1
+- `lapsedDays` 90
+- `avgCustomerValue` — default: the mean paid amount / sign-up fee in the data
+
+Hours and dollars are spread over the period the data covers (first activity
+to as-of), so a one-off $180 over one month reads as $180/month.
+
+## What comes out
+
+One JSON line (per-break count and estimate, `top3`, `totals`, `notes` for
+checks the data could not support) and, with `--out`:
+
+- `breaks.json` — everything, including every item (person label, dates,
+  amounts, systems, conflicts) and the assumptions used
+- `report.md` — the owner-facing text: top 3 breaks with count, estimate,
+  formula and what an agent could do for each, then every break, the files,
+  and the assumptions
+
+## Report back
+
+Give the owner the top 3 in plain words with the count and the estimate, and
+say that the dollars are estimates from the assumptions listed. Name people
+only when the owner needs to act on them and the exports were not hashed;
+otherwise report counts. Do not paste `breaks.json`. If `notes` says a check
+could not run (no roster export, no mailing-list export), say which export
+would make it possible.