Skills Agentes

Stripe Dispute

Evalúa si conviene económicamente contestar una disputa de Stripe, y solo si se aprueba, reúne evidencia y la envía a Stripe.

Estrellas
688

en todo el repo

Actividad
53

0–100, la ruta de este skill

Actualizado
el mes pasado

último commit aquí

Commits
1

últimos 90 días

Contexto
3.3k tok

52 tok en reposo

Paquete
1 archivo

13 KB

Instalar

Funciona con cualquier agente que lea SKILL.md

npx -y skills add OpenClaudia/openclaudia-skills --skill stripe-dispute --agent claude-code

Se instala solo en este repositorio.

Este skill makes network requests, needs API credentials.

Qué hace

  • Calcula si conviene económicamente disputar un chargeback de Stripe antes de actuar
  • Consulta el historial de disputas en Stripe para estimar la probabilidad de ganar
  • Extrae evidencia de la base de datos de usuarios (logins, uso, cancel_reason) vía queries SELECT-only
  • Genera PDFs de evidencia (política de cancelación, log de actividad) y arma el paquete
  • Sube archivos a files.stripe.com y envía la evidencia a Stripe solo tras aprobación explícita del usuario

Úsalo cuando

  • El usuario recibe una notificación de chargeback de Stripe y quiere contestarla
  • El usuario da un dispute ID (du_*) y pide 'contrarrestar'/'rebatir'/'pelear' la disputa
  • Se pide reunir evidencia para una disputa marcada fraudulent, product_not_received, product_unacceptable o subscription_canceled
  • Se quiere un PDF del log de actividad que muestre que el cliente usó el producto

No lo uses cuando

    Qué lo activa

    Di cualquiera de estas frases y el agente debería cargar este skill.

    • Recibí un chargeback en Stripe, ¿puedo contestarlo?
    • Aquí está el dispute ID du_1234, quiero rebatirlo
    • Necesito armar el paquete de evidencia para esta disputa fraudulent
    • Genera un PDF del log de actividad de este cliente para la disputa

    SKILL.md

    En inglés

    Stripe Dispute Evaluation and Evidence

    Evaluate the economics first. When contesting is justified and approved, build an evidence package and submit it to Stripe. Works for any SaaS using Stripe and a user database with login or usage logs.

    When to Use This Skill

    Use this skill when the user:

    • Receives a Stripe chargeback notification and wants to fight it
    • Provides a dispute ID (du_*) and asks to "counter" / "rebut" / "fight" it
    • Asks how to gather evidence for a dispute marked fraudulent, product_not_received, product_unacceptable, or subscription_canceled
    • Wants an activity-log PDF showing a customer used their product

    Required Environment Variables

    STRIPE_SECRET_KEY=sk_live_...      # Stripe restricted/secret key with disputes:write scope
    DATABASE_URL=postgres://...        # READ-ONLY connection to your app's user database (optional but recommended)
    TERMS_URL=https://yoursite.com/terms   # URL to your published cancellation/refund policy
    EVIDENCE_DIR=~/disputes            # Where to save the per-customer evidence folders
    

    Database safety: all queries are SELECT-only. Never let this skill issue UPDATE/DELETE/INSERT.

    Inputs

    The user provides any of:

    • Stripe dispute ID (du_xxxxx) — preferred
    • Customer email — skill will look up the dispute
    • Charge ID (ch_xxxxx or py_xxxxx)

    Decide whether to contest first

    Accepting a dispute can be the correct outcome. Never submit evidence merely because a case exists.

    1. Verify live Stripe pricing. Under current US standard pricing, Stripe charges a $15 dispute-received fee automatically and another $15 dispute-countered fee when evidence is submitted. The received fee is sunk. The countered fee is returned only on a win. Use current live pricing if these amounts change.
    2. Query the merchant's live Stripe dispute history. For defended win rate, count only closed disputes with evidence_details.submission_count > 0. Separate initial purchases from renewals and narrow to the same reason when the sample permits. Report wins, losses, open cases, and sample size. State when no exact-match precedent exists.
    3. Calculate break-even probability and expected value after the countered fee and evidence-building labor. Do not invent a probability from case strength alone.
    4. Use a conservative 30% minimum win-probability threshold unless the user provides a different threshold. If comparable data cannot support that threshold, recommend leaving the dispute unanswered. This avoids the countered fee, although the disputed amount and received fee remain lost.
    5. Obtain explicit user approval before building files, uploading evidence, or submitting. Submission is final and fee-bearing.

    Steps

    1. Pull the dispute from Stripe

    curl -s -u "$STRIPE_SECRET_KEY:" \
      "https://api.stripe.com/v1/disputes/$DISPUTE_ID" | python3 -m json.tool
    

    Extract: amount, reason, charge, evidence_details.due_by, evidence_details.submission_count, status.

    If submission_count > 0 the dispute has already been countered — STOP and warn the user.

    2. Pull the surrounding context

    # Charge → tells you the payment method, risk score, billing details, customer ID
    curl -s -u "$STRIPE_SECRET_KEY:" "https://api.stripe.com/v1/charges/$CHARGE_ID"
    
    # Customer → name, email, default payment source
    curl -s -u "$STRIPE_SECRET_KEY:" "https://api.stripe.com/v1/customers/$CUSTOMER_ID"
    
    # All invoices for the customer → look for previously-undisputed payments
    curl -s -u "$STRIPE_SECRET_KEY:" \
      "https://api.stripe.com/v1/invoices?customer=$CUSTOMER_ID&limit=100"
    
    # Subscription (if recurring)
    curl -s -u "$STRIPE_SECRET_KEY:" "https://api.stripe.com/v1/subscriptions/$SUB_ID"
    

    Prior undisputed payments on the same card are useful corroboration for fraudulent claims. Always count them, but do not treat them as proof by themselves.

    3. Look up the customer in your app database

    Adapt these queries to your schema. A useful evidence shape:

    -- User profile and self-reported cancel reason
    SELECT id, email, created_at, plan_tier, stripe_customer_id,
           cancel_reason, cancelled_at, delete_reason
    FROM users WHERE email ILIKE :email;
    
    -- Login activity (timestamps + country + device)
    SELECT created_at, country_code, device
    FROM user_activity WHERE user_id = :uid ORDER BY created_at;
    
    -- Things the customer created/used in your product
    SELECT name, type, created_at, updated_at
    FROM projects WHERE user_id = :uid AND deleted = false ORDER BY created_at;
    
    -- Checkout / payment-related actions (proves intent)
    SELECT timestamp, endpoint, payload FROM action_logs
    WHERE user_id = :uid
      AND endpoint ~* '(subscribe|checkout|stripe|upgrade|pay)'
    ORDER BY timestamp DESC;
    

    For product_not_received claims, check the user's self-reported cancel_reason. If they cancelled citing "Poor user experience" or anything admitting use, that field directly contradicts the claim. Quote it verbatim, but do not infer a win rate from one fact.

    4. Download supporting documents

    FOLDER="$EVIDENCE_DIR/$(echo $CUSTOMER_NAME | tr '[:upper:] ' '[:lower:]-')-$(date +%Y-%m)"
    mkdir -p "$FOLDER"
    
    # Invoice PDFs (URLs come from the Stripe invoice objects)
    curl -sL "$INVOICE_PDF_URL" -o "$FOLDER/invoice.pdf"
    

    5. Capture your terms / cancellation policy as a PDF

    Using Playwright (Node):

    node -e "
    const { chromium } = require('playwright');
    (async () => {
      const browser = await chromium.launch();
      const page = await browser.newPage({ viewport: { width: 1280, height: 900 } });
      await page.goto(process.env.TERMS_URL, { waitUntil: 'networkidle' });
      await page.pdf({ path: process.argv[1], format: 'A4', printBackground: true });
      await browser.close();
    })();
    " "$FOLDER/cancellation_policy.pdf"
    

    6. Generate the activity-log PDF

    Build a concise, reviewable HTML activity log, then convert it to PDF with WeasyPrint:

    from weasyprint import HTML
    HTML('activity_log.html').write_pdf('activity_log.pdf')
    

    HTML must include <meta charset="UTF-8"> to avoid mangled characters in customer names/addresses.

    A reviewable layout:

    1. Summary grid — Customer / Email / Internal user ID / Stripe customer ID / Account created / Plan / Subscription start / Cancellation timestamp (with delta from signup) / Self-reported cancel reason / Credits or units consumed / Purchase IP / Billing address / Payment method (last4, brand) / Stripe risk assessment / Referral source / Payment trigger
    2. Payment History table — every invoice with date, amount, status. Highlight the disputed row in red. Add a column "Disputed?" with explicit Yes/No.
    3. Checkout Attempts table — proves deliberate purchase intent (defeats fraudulent by extension)
    4. Items Created / Used table — names, types, timestamps. Concrete usage > adjectives.
    5. Login Activity Log — every session: timestamp, country, device. Consistency = same cardholder.
    6. Timeline Summary — single chronological narrative ending with "and then on $DATE the dispute was filed."
    7. Conclusion paragraph — quote the user's own cancel_reason against the stated dispute reason, if they contradict.

    Prefer concrete numbers over adjectives. "29 login sessions, 3 named projects, 29,960 credits consumed" is more useful than "used extensively."

    7. Show everything to the user before submitting

    open "$FOLDER"
    

    Display the rebuttal text, the evidence files, and your win-probability assessment. Wait for explicit user approval.

    8. Upload evidence files to Stripe

    Files go to files.stripe.com, NOT api.stripe.com — different host:

    curl -s -u "$STRIPE_SECRET_KEY:" \
      -F "purpose=dispute_evidence" \
      -F "file=@$FOLDER/activity_log.pdf" \
      https://files.stripe.com/v1/files
    

    Returns {"id": "file_xxx", ...}. Capture the id — that's what you reference in evidence fields.

    One file_id per dispute. Stripe rejects with 400 "That file is already attached to something else" if you try to reuse a file_id across disputes (especially for service_documentation). When fighting N disputes for the same customer, upload N copies of every shared PDF — same content, fresh file_id each time.

    9. Submit evidence (one shot — submit=true is final)

    curl -s -u "$STRIPE_SECRET_KEY:" \
      -X POST "https://api.stripe.com/v1/disputes/$DISPUTE_ID" \
      -d "evidence[uncategorized_text]=$REBUTTAL_TEXT" \
      -d "evidence[uncategorized_file]=$ACTIVITY_LOG_FILE_ID" \
      -d "evidence[receipt]=$INVOICE_FILE_ID" \
      -d "evidence[cancellation_policy]=$TERMS_FILE_ID" \
      -d "evidence[cancellation_policy_disclosure]=$CANCEL_DISCLOSURE_TEXT" \
      -d "evidence[refund_policy]=$TERMS_FILE_ID" \
      -d "evidence[refund_policy_disclosure]=$REFUND_DISCLOSURE_TEXT" \
      -d "evidence[cancellation_rebuttal]=$CANCEL_REBUTTAL_TEXT" \
      -d "evidence[access_activity_log]=$ACCESS_LOG_SUMMARY" \
      -d "evidence[service_date]=$SERVICE_START_DATE" \
      -d "evidence[product_description]=$PRODUCT_DESCRIPTION" \
      -d "evidence[customer_email_address]=$CUSTOMER_EMAIL" \
      -d "evidence[customer_name]=$CUSTOMER_NAME" \
      -d "evidence[customer_purchase_ip]=$PURCHASE_IP" \
      -d "evidence[billing_address]=$BILLING_ADDRESS" \
      -d "submit=true" \
      "https://api.stripe.com/v1/disputes/$DISPUTE_ID"
    

    Verify the response:

    • status should be under_review
    • evidence_details.has_evidence should be true
    • evidence_details.submission_count should be 1

    Evidence Strategy by Dispute Reason

    fraudulent

    Goal: prove the cardholder made the purchase.

    • Prior undisputed payments on the same card (strongest)
    • OAuth login (Google/Apple) = identity-verified signup
    • Consistent IP / country / device across sessions
    • Multiple checkout attempts before purchase = real person deliberating
    • Stripe's own risk assessment was normal
    • Subscription still active and not cancelled

    product_not_received

    Goal: prove delivery + use.

    • Login activity log
    • Items created / actions taken inside the product
    • The customer's own self-reported cancel reason, if they cancelled — quoted verbatim against the dispute claim
    • Receipt and welcome email

    product_unacceptable

    Goal: show the product matched its description and the customer used it.

    • Same as product_not_received PLUS your terms-of-service language about quality / refund policy
    • Highlight that customer never opened a support ticket

    subscription_canceled

    Goal: prove the customer never cancelled (or cancelled after the renewal).

    • Subscription object showing cancel_at_period_end=false at the renewal date
    • All login/use activity from after the renewal date
    • Terms language stating annual renewals require explicit cancellation

    Rebuttal Templates

    uncategorized_text

    [Customer name] created a [Product name] account on [date] via [auth method] and subscribed to [plan] ($[amount]/[interval]) using the same [card brand]. The first [N] payment(s) were never disputed. The customer actively used the service: [N] login sessions from [country] on [device], [N] items created ([list]), and [N] [units] consumed. The disputed charge is the [renewal/initial] payment on [date]. The subscription was [status] and remains [active/cancelled]. The customer never contacted support to cancel or request a refund. Our cancellation and refund policies are published at [TERMS_URL]. This is not a fraudulent transaction — it is a legitimate purchase from the cardholder who [made/has made] [N] other undisputed payments on this account.

    cancellation_policy_disclosure

    Our cancellation policy is disclosed at [TERMS_URL]. Subscribers may cancel at any time and retain access through the end of their billing cycle. This customer never cancelled.

    refund_policy_disclosure

    Our refund policy is disclosed at [TERMS_URL]. We offer a [N]-day money-back guarantee. The customer did not request a refund within that window, nor at any time.

    Important Notes

    • File uploads go to files.stripe.com, NOT api.stripe.com
    • One submission per dispute. submit=true is final
    • Evidence deadline is evidence_details.due_by (unix timestamp). After that you can no longer submit
    • Database queries must be READ-ONLY — restrict the connection's role if possible
    • Never include other customers' data in the activity log PDF
    • Save every evidence package to disk in case you need to reference it for future disputes from the same customer

    High-value contradiction pattern

    Customer signs up, uses product briefly, cancels within hours citing "Poor user experience" in your in-app cancel form, then files a chargeback days later claiming "product not received."

    The cancel form's reason directly contradicts the chargeback claim. Quote it word-for-word in the rebuttal, while still applying the fee and historical-outcome decision gate.

    Always check users.cancel_reason (or your equivalent) FIRST when the dispute reason is product_not_received or product_unacceptable.

    Reproducido de OpenClaudia/openclaudia-skills bajo licencia MIT. Leer esta página en markdown.

    Archivos

    1 archivo en el paquete. Solo se lee SKILL.md al activarse — las referencias se cargan si el skill decide que las necesita.

    Antes de instalar

    Requiere STRIPE_SECRET_KEY con scope disputes:write, opcionalmente DATABASE_URL de solo lectura, TERMS_URL y EVIDENCE_DIR, además de Playwright y WeasyPrint para generar PDFs.

    Necesita en el PATH:curlnodepython3

    Variables de entorno:ACCESS_LOG_SUMMARYACTIVITY_LOG_FILE_IDBILLING_ADDRESSCANCEL_DISCLOSURE_TEXTCANCEL_REBUTTAL_TEXTCHARGE_IDCUSTOMER_EMAILCUSTOMER_IDCUSTOMER_NAMEDISPUTE_IDEVIDENCE_DIRFOLDERINVOICE_FILE_IDINVOICE_PDF_URLPRODUCT_DESCRIPTIONPURCHASE_IPREBUTTAL_TEXTREFUND_DISCLOSURE_TEXTSERVICE_START_DATESTRIPE_SECRET_KEYSUB_IDTERMS_FILE_IDTERMS_URL

    Detalles

    Categoría
    Finanzas
    Licencia
    MIT
    Recursos incluidos
    Solo SKILL.md
    Código fuente
    Ver SKILL.md

    Más de OpenClaudia/openclaudia-skills

    Este repo incluye 76 skills. Si instalas uno, normalmente ya tienes los demás. Ver el pack openclaudia-skills entero y su comando de instalación

    Genera un informe HTML autocontenible de tráfico competitivo: visitas mensuales (SimilarWeb), tráfico orgánico y Domain Rating (Ahrefs), con gráficos ranked, tendencias y tabla de datos.

    Costo de contexto al activarse
    1.2k tok
    Tamaño del paquete
    4 archivos
    Última actualización
    hace 4 días
    seo geo

    Obtiene datos de marca (nombre, descripción, logos, industria) desde la API de brand.dev y guarda los logos localmente.

    Costo de contexto al activarse
    702 tok
    Tamaño del paquete
    1 archivo
    Última actualización
    el mes pasado
    datos analitica

    Audita TODAS las propiedades de Google Search Console a la vez: ranking por clics e impresiones con deltas, y diff de keywords por sitio (nuevas, suben, bajan, perdidas, o bien rankeadas sin clics).

    Costo de contexto al activarse
    1k tok
    Tamaño del paquete
    2 archivos
    Última actualización
    el mes pasado
    seo geo

    Clasifica y resume el feed de WeChat Moments (朋友圈) del usuario para que los eventos reales y la información genuina destaquen sobre la promoción, ponderando según cuánto le escribe el usuario a cada autor.

    Costo de contexto al activarse
    1.2k tok
    Tamaño del paquete
    1 archivo
    Última actualización
    el mes pasado
    redes sociales

    Edita audio o video de podcast: recorta charla previa/posterior, quita muletillas, corta silencios, mejora el audio y aplica el mismo corte a una versión en video.

    Costo de contexto al activarse
    6k tok
    Tamaño del paquete
    3 archivos
    Última actualización
    el mes pasado
    redaccion contenido

    Genera un informe de Citaciones de IA (GEO) para un dominio: qué prompts de búsqueda con IA citan el sitio en Google AI Overview y ChatGPT, con contexto de tráfico orgánico y cobertura por artículo.

    Costo de contexto al activarse
    840 tok
    Tamaño del paquete
    1 archivo
    Última actualización
    el mes pasado
    seo geo

    Skills relacionados

    Consulta la API REST U.S. Treasury Fiscal Data para datos financieros federales de EE. UU., sin API key: deuda nacional, estados diarios y mensuales del Tesoro, subastas de valores, tasas de interés, tipo de cambio y bonos de ahorro.

    Costo de contexto al activarse
    1.7k tok
    Tamaño del paquete
    9 archivos
    Última actualización
    el mes pasado
    finanzas

    Marco de pensamiento de Charlie Munger, destilado de Poor Charlie's Almanack y juntas de accionistas: modelos mentales, heurísticas y estilo, para revisar decisiones de inversión y detectar sesgos cognitivos.

    Costo de contexto al activarse
    7.5k tok
    Tamaño del paquete
    6 archivos
    Última actualización
    hace 2 meses
    finanzas

    Crea órdenes de compra aprobadas usando la tool `create-purchase-order`.

    Costo de contexto al activarse
    99 tok
    Tamaño del paquete
    3 archivos
    Última actualización
    el mes pasado
    Oficialfinanzas