Respuesta rápida
- Qué es
- crm-data-cleanup es una skill de OneWave AI (licencia MIT) que le enseña a Claude a limpiar la exportación de cualquier CRM: encuentra duplicados con coincidencia aproximada, normaliza datos y arma un plan de fusión que aprueba una persona, sin tocar el CRM.
- Qué le das
- La exportación en CSV de contactos (y de empresas, si quieres) con los identificadores de cada registro
- Qué te devuelve
- Un plan de fusión revisable, la cola de pares dudosos, la lista de registros marcados y archivos limpios listos para importar
- Autor
- OneWave AI
- Licencia
- MIT
- Repositorio
- onewave-ai/claude-skills
- En GitHub
- 430 estrellas · actualizada el
- Revisada por Claura
Instálala en una línea para todos tus proyectos, en Claude Code, Cursor, Codex y más:
npx skills add onewave-ai/claude-skills --skill crm-data-cleanup -g
Funciona en
~/.claude/skills/~/.agents/skills/~/.agents/skills/~/.agents/skills/Pedidos para probarla
Con la skill instalada, copia uno y pégaselo a tu agente tal cual, o cambia los datos por los tuyos.
Limpia esta exportación de contactos del CRM [adjunta el CSV]: duplicados, mails mal escritos y teléfonos sin formato, y dame el archivo listo para reimportar.
Encuentra las empresas duplicadas en mi lista aunque estén escritas distinto (S.A., SA, mayúsculas).
Dame un informe de calidad de datos de mi CRM: qué porcentaje de contactos no tiene mail, cargo o empresa.
#Qué hace
crm-data-cleanup es una skill de OneWave AI (licencia MIT) que convierte una exportación desordenada del CRM en un plan de fusión que una persona puede revisar, más archivos listos para importar. Parte de tres fallas típicas: comparar solo correos exactos deja pasar la mayoría de los duplicados, fusionar por buzones compartidos como info@ junta a personas distintas, y una fusión puede pisar datos buenos con vacíos sin vuelta atrás. Por eso la skill nunca fusiona: planifica, una persona aprueba y la fusión la hace la herramienta del CRM, que es la que conserva el historial de actividad.
El trabajo lo hace scripts/dedupe.py, un script de Python que no llama a ninguna API ni modifica los archivos de entrada. Detecta las columnas por los encabezados habituales de cada proveedor, normaliza correos, teléfonos al formato internacional E.164, nombres, dominios y países, agrupa candidatos, les da un puntaje y elige qué registro sobrevive (por defecto, el de actividad más reciente, después el más completo, después el más antiguo). Sus reglas de seguridad: una fusión automática exige un correo o teléfono único compartido más un nombre compatible; un nombre idéntico en la misma empresa solo va a revisión; cada par dentro de un grupo se vuelve a comprobar, y los grupos de más de cinco registros pasan a revisión. Juntar variantes de Gmail con puntos o con + es opcional, porque en dominios de empresa esas direcciones son distintas.
Entrega merge_plan.csv (una fila por grupo y campo, con el valor que gana y de dónde sale), review_queue.csv (pares para decidir a mano), cleaned_contacts.csv, record_flags.csv (nueve marcas, como correo inválido, error de dominio, buzón de rol, registro de prueba o sin actividad en un año), associations.csv (empresas huérfanas o contactos sin empresa) y summary.json. Incluye evaluate.py para medir aciertos contra un archivo de verdad, datos de prueba con 41 contactos y 14 duplicados reales con trampas sembradas, y tres referencias: estándares de campos, notas por CRM sobre qué destruye cada fusión, y un plan de despliegue seguro en siete fases con respaldo y vuelta atrás.
#Cómo aplicarla a tu negocio
Caso de ejemplo: una inmobiliaria en Madrid que va a pasar de Zoho a otro CRM con unos 12.000 contactos acumulados entre portales, formularios y ferias. Antes de migrar quiere sacar los duplicados.
- Respaldo primero. Exporta contactos y empresas con identificadores y asociaciones, y guarda esa copia con fecha. La skill lo pide antes de cualquier fusión porque es la única vuelta atrás real.
- Corre el ensayo con la región correcta. El agente ejecuta el script con
--default-region ESy--as-ofcon la fecha de hoy, para que los teléfonos españoles se normalicen bien y la corrida se pueda repetir. - Revisa el resumen antes de creer un número. Si la columna «Etapa» de negocios se tomó como etapa del contacto, se corrige con
--coly se vuelve a correr. - Recorre la cola de revisión con quien conoce la base. Por ejemplo, 180 pares dudosos: dos asesores con el mismo apellido, una pareja que comparte correo, una oficina con un solo teléfono. Cada par queda como fusionar o mantener.
- Fusiona de a poco y después importa. Diez grupos de prueba en la herramienta del CRM, verificación y el resto por tandas. Recién entonces se importan los archivos limpios, enlazados por identificador, al CRM nuevo.
#Cómo instalarla
La skill corre en cualquier agente que pueda ejecutar Python. Trae scripts, datos de prueba y referencias, así que hay que instalar la carpeta entera, no solo el SKILL.md.
- Con una línea, para cualquier agente:
npx skills add onewave-ai/claude-skills --skill crm-data-cleanup -gla deja instalada en todos tus proyectos. - A mano: copia la carpeta
crm-data-cleanup/completa (conscripts/,assets/yreferences/) a~/.claude/skills/para Claude Code, o a~/.agents/skills/para Cursor, Codex y Gemini CLI. - App de Claude: comprime la carpeta en un ZIP y súbela desde Customize, Skills, «Upload a skill».
- Qué más necesitas: Python con
pip install pandas rapidfuzz phonenumbers(el último es opcional pero recomendado). No pide claves ni acceso al CRM: trabaja solo sobre archivos CSV.
El SKILL.md completo (con la licencia y la fuente) está más abajo.
El SKILL.md completo
Skill original de OneWave AI (MIT). Fuente: github.com/onewave-ai/claude-skills/tree/main/crm-data-cleanup.
El archivo original está en inglés y va tal cual, sin traducir. Lo de arriba de esta ficha es el resumen en español: qué hace, cómo aplicarla y cómo instalarla.
Leer el archivo (154 líneas)
CRM Data Cleanup
This skill turns a messy CRM export into a merge plan a human can check, plus files that are ready to import. It works the same way for any vendor, because every CRM exports CSV.
Duplicate cleanup usually fails in one of three ways. Exact-email matching misses most
duplicates. Merging on shared inboxes like info@ fuses different people into one. And a
merge overwrites good values with blanks, losing history that can't be recovered. Most
CRM merges can't be undone, so this skill never merges anything itself. It plans, a
human approves, and the CRM's own merge tool does the merging.
Files
scripts/dedupe.py: the dry-run engine: normalize, block, score, cluster, pick a survivor, merge fields. Needspandasandrapidfuzz. Usesphonenumbersif it's installed.scripts/evaluate.py: scores a run against a labeled truth file (pairwise precision and recall).assets/fixture/: 41 synthetic contacts and 14 companies with planted duplicates, plustruth.csv.references/field-standards.md: normalization rules, state and country codes, lifecycle stage mapping, and flag meanings.references/vendor-notes.md: what merge keeps and destroys in each CRM, import quirks, and links to the official docs.references/safe-rollout.md: backup, sample merge, batch execution, and rollback.
Workflow
- Get the exports. Ask for a full-property CSV of contacts (or leads/people) and, when companies are in scope, companies (accounts/organizations) with record IDs. Record IDs matter because the cleaned files update records by ID.
- Back up first. Before any merge, confirm the user has a dated full export that
includes associations (
safe-rollout.md, Phase 0). A backup is the only real rollback. - Run the dry run:
The script detects columns from common vendor headers. Fix a wrong guess with
pip install pandas rapidfuzz phonenumbers # phonenumbers is optional but preferred python scripts/dedupe.py --contacts contacts.csv --companies companies.csv \ --out plan/ --as-of 2026-09-22 [--default-region GB] [--gmail-canonical]--col email="E-mail 1"(or--company-colfor the companies file). - Check
summary.jsonbefore trusting any number. Confirmcolumn_map(a dealStagecolumn mapped as lifecycle skews everything),phone_engine, andwarnings, such as a block that was skipped for being too large. - Present the results by tier. Auto clusters: the count and a few examples. Review queue: every pair, with its reasons. Flags: counts by type. Associations: orphaned and missing links. Lead with the review queue, because it's where the user's judgment is needed.
- Walk the review queue with the data owner. Record
mergeorkeepin thedecisioncolumn. Then look at thekept_survivor_conflictrows inmerge_plan.csv: those are the values that will disappear. - Hand off the execution plan. Merge a sample of 5 to 10 clusters in the native tool,
verify, then do the rest in batches. After merging, import the cleaned files keyed on
record ID. Follow
safe-rollout.md, and checkvendor-notes.mdfor the specific CRM.
Merge-safety rules
- Never auto-merge on a role or shared identifier.
info@,sales@, a household inbox, or an office main line shows that records belong to the same organization, not that they're the same person. A shared identifier goes to review at most. - Auto-merge needs a strong ID plus a compatible name. A strong ID is an exact non-shared email, or an exact non-shared phone. A name match alone, even 100% with the same company, only goes to review, because two people named David Kim at one company is normal in large organizations.
- Re-check every pair inside a cluster. A=B by email and B=C by phone can chain two
different people together. The script demotes any cluster with an incompatible pair,
or with more than
--max-cluster(5) members, to review. - Never overwrite a non-empty value with an empty one. The survivor's value wins.
Blanks are filled from the best-ranked loser. Create date takes the earliest, last
activity the latest, and lifecycle stage the furthest along. Losing emails and phones
go to
additional_emailsandadditional_phones, so they aren't dropped. - Pick the survivor on purpose. The default order is
most_recent_activity, thenmost_complete, thenoldest_created. Addbusiness_emailfirst when the work address should be the primary email. In HubSpot and Pipedrive the primary's values win every conflict, so the survivor choice decides what's kept. - Gmail dot and plus folding is opt-in only (
--gmail-canonical). Google says dots don't matter for personal Gmail, but they do matter for Workspace domains. Applied by default, it would merge real, distinct mailboxes. - Stay a dry run. Never call a CRM API, and never write to the input files. A merge runs in the CRM's native tool, which moves activity history. Delete-and-reimport doesn't.
- Flag, don't delete. Stale, test, and invalid records go in
record_flags.csv. The data owner decides whether to archive or delete them.
Output format
| File | Contents |
|---|---|
merge_plan.csv | One row per cluster per field: cluster_id, survivor_id, loser_ids, survivor_rule, field, winning_value, source_id, action, other_values. The action is kept_survivor, kept_survivor_conflict, filled_from_loser, earliest, latest, furthest_stage, or preserved_secondary |
review_queue.csv | Pairs needing a human: both records side by side, confidence, reasons, suggested_survivor, and blank decision and decided_by columns |
cleaned_contacts.csv | The original columns with normalized values. Survivors carry merged values, and losers are removed. Also additional_emails, additional_phones, _cleanup_action, and _flags |
record_flags.csv | invalid_email, email_domain_typo, role_email, shared_phone, no_contact_method, phone_unparsed, junk_or_test, missing_name, stale |
companies_*.csv, cleaned_companies.csv | The same outputs for companies (domain-first matching) |
associations.csv | company_merged (the ID was re-pointed to the survivor), orphan_company_id, missing_association (the email domain matches a company) |
summary.json | Counts, flag totals, column map, settings, warnings |
When reporting to the user, give: records in and out, auto clusters, review pairs, flag
counts, the three riskiest review pairs, and the next gate from safe-rollout.md.
Worked example (bundled fixture)
python scripts/dedupe.py --contacts assets/fixture/contacts.csv \
--companies assets/fixture/companies.csv --out plan/ --as-of 2026-09-22
python scripts/evaluate.py --plan-dir plan/ --truth assets/fixture/truth.csv
The fixture is 41 contacts holding 14 true duplicate pairs. It includes case and
whitespace email variants, a name typo, a nickname with a reformatted phone, the same
person with a work and a Gmail address, Inc versus no Inc, a hyphenated married name,
a 3-record cluster, and a Gmail dot/plus variant. It also plants traps that must not
merge: three people sharing info@stark.com, a couple sharing a household inbox, three
colleagues on one office line, and two different people named David Kim.
| Run | Auto pairs | Wrong auto | Auto precision | Auto recall | Recall incl. review |
|---|---|---|---|---|---|
| Naive: exact lowercase email only | 11 | 4 (the info@ trio and the household) | 63.6% | 50.0% | n/a |
| Default | 11 | 0 | 100% | 78.6% | 100% |
--gmail-canonical | 12 | 0 | 100% | 85.7% | 100% |
In the default run, 7 contact clusters absorb 9 records, and 6 pairs go to review: 3 true
duplicates plus the household and info@ traps. For companies, Acme Inc/ACME and
Tyrell Corp/The Tyrell Corporation auto-merge by domain. Globex LLC/Globex Corporation (no domain) and Umbrella Ltd/Umbrella Corporation (different domains) go
to review. Three association issues surface, including an orphaned CO99 reference.
Cluster MC0001 (Jane Doe x3) shows the field rules at work. Create date 2022-03-14
comes from the oldest record. Lifecycle stage customer is the furthest along. The blank
state on the survivor is filled from C001, and the work email is kept in
additional_emails.
Tuning
--auto-name-sim 85,--review-name-sim 88,--conflict-name-sim 60: raise the auto threshold for large, name-dense databases.--stale-days 365,--as-of: set the staleness window. Always pass--as-ofso reruns are reproducible.--max-block 500: blocks bigger than this are skipped with a warning. Keeping blocks small keeps pair scoring fast on 100k-row exports.- Extend
NICKNAMES,ROLE_LOCALS,LIFECYCLE_MAP, orLEGAL_SUFFIXESin the script for a specific region or portal. Custom HubSpot lifecycle stages have numeric internal IDs.
Copiar el archivo tal cual
---
name: crm-data-cleanup
description: Cleans up CRM data from HubSpot, Salesforce, Pipedrive, Zoho, Close, Attio, or any CRM CSV export. Finds duplicate contacts and companies with fuzzy matching (not just exact email), normalizes emails, phones to E.164, names, domains, states, and countries, flags stale, invalid, role, and test records, finds broken contact-company associations, and produces a reviewable merge plan plus import-ready files without touching the live CRM. Use this skill whenever the user wants to dedupe, find or merge duplicate contacts, merge companies or accounts, clean up a CRM, prep a CRM migration or re-import, or says the CRM export is a mess. Also use it when they mention only one symptom, like "the same person is in here three times", "half our leads are junk", or "our company records don't match".
---
<!--
Skill original: crm-data-cleanup, de OneWave AI (https://github.com/onewave-ai/claude-skills/tree/main/crm-data-cleanup)
Licencia: MIT. Reproducida con atribución según los términos de la licencia.
Copia publicada en claura-ai.com/skills.
-->
# CRM Data Cleanup
This skill turns a messy CRM export into a merge plan a human can check, plus files that
are ready to import. It works the same way for any vendor, because every CRM exports CSV.
Duplicate cleanup usually fails in one of three ways. Exact-email matching misses most
duplicates. Merging on shared inboxes like `info@` fuses different people into one. And a
merge overwrites good values with blanks, losing history that can't be recovered. Most
CRM merges **can't be undone**, so this skill never merges anything itself. It plans, a
human approves, and the CRM's own merge tool does the merging.
## Files
- `scripts/dedupe.py`: the dry-run engine: normalize, block, score, cluster, pick a
survivor, merge fields. Needs `pandas` and `rapidfuzz`. Uses `phonenumbers` if it's
installed.
- `scripts/evaluate.py`: scores a run against a labeled truth file (pairwise precision
and recall).
- `assets/fixture/`: 41 synthetic contacts and 14 companies with planted duplicates,
plus `truth.csv`.
- `references/field-standards.md`: normalization rules, state and country codes,
lifecycle stage mapping, and flag meanings.
- `references/vendor-notes.md`: what merge keeps and destroys in each CRM, import
quirks, and links to the official docs.
- `references/safe-rollout.md`: backup, sample merge, batch execution, and rollback.
## Workflow
1. **Get the exports.** Ask for a full-property CSV of contacts (or leads/people) and,
when companies are in scope, companies (accounts/organizations) with record IDs.
Record IDs matter because the cleaned files update records by ID.
2. **Back up first.** Before any merge, confirm the user has a dated full export that
includes associations (`safe-rollout.md`, Phase 0). A backup is the only real rollback.
3. **Run the dry run:**
```bash
pip install pandas rapidfuzz phonenumbers # phonenumbers is optional but preferred
python scripts/dedupe.py --contacts contacts.csv --companies companies.csv \
--out plan/ --as-of 2026-09-22 [--default-region GB] [--gmail-canonical]
```
The script detects columns from common vendor headers. Fix a wrong guess with
`--col email="E-mail 1"` (or `--company-col` for the companies file).
4. **Check `summary.json` before trusting any number.** Confirm `column_map` (a deal
`Stage` column mapped as lifecycle skews everything), `phone_engine`, and `warnings`,
such as a block that was skipped for being too large.
5. **Present the results by tier.** Auto clusters: the count and a few examples. Review
queue: every pair, with its reasons. Flags: counts by type. Associations: orphaned
and missing links. Lead with the review queue, because it's where the user's judgment
is needed.
6. **Walk the review queue with the data owner.** Record `merge` or `keep` in the
`decision` column. Then look at the `kept_survivor_conflict` rows in
`merge_plan.csv`: those are the values that will disappear.
7. **Hand off the execution plan.** Merge a sample of 5 to 10 clusters in the native tool,
verify, then do the rest in batches. After merging, import the cleaned files keyed on
record ID. Follow `safe-rollout.md`, and check `vendor-notes.md` for the specific CRM.
## Merge-safety rules
- **Never auto-merge on a role or shared identifier.** `info@`, `sales@`, a household
inbox, or an office main line shows that records belong to the same organization, not
that they're the same person. A shared identifier goes to review at most.
- **Auto-merge needs a strong ID plus a compatible name.** A strong ID is an exact
non-shared email, or an exact non-shared phone. A name match alone, even 100% with the
same company, only goes to review, because two people named David Kim at one company is
normal in large organizations.
- **Re-check every pair inside a cluster.** A=B by email and B=C by phone can chain two
different people together. The script demotes any cluster with an incompatible pair,
or with more than `--max-cluster` (5) members, to review.
- **Never overwrite a non-empty value with an empty one.** The survivor's value wins.
Blanks are filled from the best-ranked loser. Create date takes the earliest, last
activity the latest, and lifecycle stage the furthest along. Losing emails and phones
go to `additional_emails` and `additional_phones`, so they aren't dropped.
- **Pick the survivor on purpose.** The default order is `most_recent_activity`, then
`most_complete`, then `oldest_created`. Add `business_email` first when the work
address should be the primary email. In HubSpot and Pipedrive the primary's values win
every conflict, so the survivor choice decides what's kept.
- **Gmail dot and plus folding is opt-in only** (`--gmail-canonical`). Google says dots
don't matter for personal Gmail, but they do matter for Workspace domains. Applied by
default, it would merge real, distinct mailboxes.
- **Stay a dry run.** Never call a CRM API, and never write to the input files. A merge
runs in the CRM's native tool, which moves activity history. Delete-and-reimport
doesn't.
- **Flag, don't delete.** Stale, test, and invalid records go in `record_flags.csv`. The
data owner decides whether to archive or delete them.
## Output format
| File | Contents |
|---|---|
| `merge_plan.csv` | One row per cluster per field: `cluster_id, survivor_id, loser_ids, survivor_rule, field, winning_value, source_id, action, other_values`. The `action` is `kept_survivor`, `kept_survivor_conflict`, `filled_from_loser`, `earliest`, `latest`, `furthest_stage`, or `preserved_secondary` |
| `review_queue.csv` | Pairs needing a human: both records side by side, `confidence`, `reasons`, `suggested_survivor`, and blank `decision` and `decided_by` columns |
| `cleaned_contacts.csv` | The original columns with normalized values. Survivors carry merged values, and losers are removed. Also `additional_emails`, `additional_phones`, `_cleanup_action`, and `_flags` |
| `record_flags.csv` | `invalid_email`, `email_domain_typo`, `role_email`, `shared_phone`, `no_contact_method`, `phone_unparsed`, `junk_or_test`, `missing_name`, `stale` |
| `companies_*.csv`, `cleaned_companies.csv` | The same outputs for companies (domain-first matching) |
| `associations.csv` | `company_merged` (the ID was re-pointed to the survivor), `orphan_company_id`, `missing_association` (the email domain matches a company) |
| `summary.json` | Counts, flag totals, column map, settings, warnings |
When reporting to the user, give: records in and out, auto clusters, review pairs, flag
counts, the three riskiest review pairs, and the next gate from `safe-rollout.md`.
## Worked example (bundled fixture)
```bash
python scripts/dedupe.py --contacts assets/fixture/contacts.csv \
--companies assets/fixture/companies.csv --out plan/ --as-of 2026-09-22
python scripts/evaluate.py --plan-dir plan/ --truth assets/fixture/truth.csv
```
The fixture is 41 contacts holding 14 true duplicate pairs. It includes case and
whitespace email variants, a name typo, a nickname with a reformatted phone, the same
person with a work and a Gmail address, `Inc` versus no `Inc`, a hyphenated married name,
a 3-record cluster, and a Gmail dot/plus variant. It also plants traps that must not
merge: three people sharing `info@stark.com`, a couple sharing a household inbox, three
colleagues on one office line, and two different people named David Kim.
| Run | Auto pairs | Wrong auto | Auto precision | Auto recall | Recall incl. review |
|---|---|---|---|---|---|
| Naive: exact lowercase email only | 11 | 4 (the `info@` trio and the household) | 63.6% | 50.0% | n/a |
| Default | 11 | 0 | 100% | 78.6% | 100% |
| `--gmail-canonical` | 12 | 0 | 100% | 85.7% | 100% |
In the default run, 7 contact clusters absorb 9 records, and 6 pairs go to review: 3 true
duplicates plus the household and `info@` traps. For companies, `Acme Inc`/`ACME` and
`Tyrell Corp`/`The Tyrell Corporation` auto-merge by domain. `Globex LLC`/`Globex
Corporation` (no domain) and `Umbrella Ltd`/`Umbrella Corporation` (different domains) go
to review. Three association issues surface, including an orphaned `CO99` reference.
Cluster `MC0001` (Jane Doe x3) shows the field rules at work. Create date `2022-03-14`
comes from the oldest record. Lifecycle stage `customer` is the furthest along. The blank
state on the survivor is filled from C001, and the work email is kept in
`additional_emails`.
## Tuning
- `--auto-name-sim 85`, `--review-name-sim 88`, `--conflict-name-sim 60`: raise the auto
threshold for large, name-dense databases.
- `--stale-days 365`, `--as-of`: set the staleness window. Always pass `--as-of` so reruns
are reproducible.
- `--max-block 500`: blocks bigger than this are skipped with a warning. Keeping blocks
small keeps pair scoring fast on 100k-row exports.
- Extend `NICKNAMES`, `ROLE_LOCALS`, `LIFECYCLE_MAP`, or `LEGAL_SUFFIXES` in the script
for a specific region or portal. Custom HubSpot lifecycle stages have numeric internal
IDs.
Preguntas frecuentes sobre la skill crm-data-cleanup
¿Qué es la skill crm-data-cleanup y para qué sirve?
crm-data-cleanup es una skill de OneWave AI (github.com/onewave-ai/claude-skills, licencia MIT) que limpia exportaciones de HubSpot, Salesforce, Pipedrive, Zoho, Close, Attio o cualquier CRM que exporte CSV. Trae un script en Python que normaliza correos, teléfonos, nombres, dominios y países, encuentra duplicados con coincidencia aproximada (no solo por correo idéntico), marca registros de prueba, viejos o inválidos y detecta contactos mal asociados a empresas. Nunca fusiona nada: entrega un plan que una persona revisa y después ejecuta con la herramienta de fusión del propio CRM.
¿Cómo la aplico a mi base de contactos?
Exporta contactos y empresas con todas las propiedades y los identificadores de registro, y guarda una copia fechada como respaldo. Pídele al agente que corra el ensayo y revise primero summary.json, donde se ve qué columna tomó como correo, teléfono o etapa. Después recorres con quien conoce la base la cola de revisión (review_queue.csv), marcando fusionar o mantener en cada par. Fusionas a mano entre 5 y 10 grupos de prueba, verificas, y sigues por tandas. Si tu base nace de formularios mal cargados, lee también cuándo conviene automatizar la carga de leads en tu CRM.
¿En qué se diferencia de la función de duplicados que ya trae mi CRM?
La mayoría de los CRM compara correos exactos, y eso deja pasar la mitad de los duplicados y además junta por error a personas distintas que comparten un buzón como info@. En los datos de prueba que trae la skill, comparar solo por correo idéntico acierta el 63,6 % de las fusiones automáticas; el modo por defecto del script acierta el 100 % y manda a revisión humana los casos dudosos. También evita perder datos: nunca reemplaza un valor lleno por uno vacío, se queda con la fecha de creación más antigua y la etapa más avanzada, y guarda los correos y teléfonos de los registros absorbidos.
¿Qué necesito instalado y sirve para bases en español?
Necesitas Python con pandas y rapidfuzz, y conviene sumar phonenumbers para validar teléfonos. La skill está en inglés y su script viene pensado para Estados Unidos: los teléfonos se interpretan como de ese país salvo que pases --default-region con tu código (MX, AR, CO, ES), y las listas de apodos y de buzones compartidos están en inglés. Puedes pedirle al agente que agregue Pepe por José o buzones como ventas@ y contacto@ a esas listas antes de correrlo. Las notas por proveedor no cubren Twenty: para ese caso trae cinco preguntas a responder y una fusión de prueba.
Más skills de Ventas y CRM

email-marketing: arma newsletters, campañas y secuencias con IA (skill de Claura)
Planifica, escribe y revisa newsletters, campañas y secuencias de email con un objetivo por mail, segmentos por origen, UTMs en cada link y un control de 15 puntos antes de enviar.

ads-audit: auditar tus cuentas de anuncios con evidencia
Audita cuentas de anuncios de hasta 12 plataformas y separa lo que está bien, lo que está mal y lo que no se pudo revisar, con la evidencia de cada punto.

ads: planificar, revisar y corregir campañas en Meta y Google
Planifica, revisa y corrige campañas pagas en Google, Meta y LinkedIn, y te dice con datos si conviene cortar, mantener o escalar.
Para seguir con este tema
- Guía de Ventas y CRMEl orden para empezar y las notas de este tema.Guía
- Skills de Claude y ChatGPT para CRMLas skills de Claude para CRM: limpiar contactos duplicados, puntuar leads, revisar el pipeline y cargar los leads del correo. Cómo usarlas en Claude y en ChatGPT.Skill
- En qué momento conviene automatizar la carga de leads en tu CRMCargar leads a mano funciona hasta que deja de alcanzar. Las señales para saber cuándo automatizar la carga al CRM, qué cambia y cómo sacar la cuenta.Nota
AI KID
RE/MAX Patagon
Mateo Caño