Consumer dispute assistant
Skill JerryNee/consumer-dispute-agent-skill/skills/consumer-dispute-assistant
Build consumer dispute case packs for refunds, billing errors, subscription cancellation failures, duplicate or unauthorized charges, ecommerce/service complaints, chargeback preparation, and complaint escalation. Use when a user wants an agent to organize evidence, draft merchant support messages, prepare payment dispute statements, create regulator/BBB/state attorney general complaint drafts, or plan safe follow-up workflows. Do not use for legal advice, court representation, fabricated evidence, harassment, threats, or autonomous filing/sending without explicit user confirmation.From its SKILL.md
npx -y skills add JerryNee/consumer-dispute-agent-skill --skill consumer-dispute-assistantAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 2 commands, including `python3 scripts/build_case_pack.py intake.json --out dispute-pack --tone firm` and 1 more.
SKILL.md
6.0 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it
Consumer Dispute Assistant
Overview
Use this skill to turn messy consumer dispute materials into a factual, auditable case pack: timeline, evidence checklist, merchant message, payment dispute statement, complaint draft, and next-step plan. Keep the user in control; prepare and explain actions, but do not send messages, submit forms, initiate chargebacks, or contact third parties without explicit confirmation.
Operating Rules
- Do not present yourself as a lawyer, legal service, or guaranteed recovery tool.
- Do not provide jurisdiction-specific legal conclusions unless you have verified current official sources and cite them.
- Do not invent, embellish, forge, or imply evidence the user does not have.
- Do not threaten illegal action, reputational harm, harassment, or regulatory complaints as leverage unless the complaint is a factual next step the user can choose.
- Redact sensitive data in shareable outputs: full card numbers, SSNs, passwords, access tokens, medical details not needed for the dispute, and unnecessary addresses.
- Ask for missing facts when they materially change the action. Otherwise, mark gaps as
Needs user input. - Require user approval before any external action: sending an email, filling a web form, calling, filing a complaint, opening a chargeback, or contacting a bank.
Core Workflow
-
Classify the dispute:
- Subscription cancellation failure
- Duplicate, incorrect, or unauthorized charge
- Refund denial or return issue
- Service not provided or product not as described
- Travel, lodging, telecom, utility, banking, or marketplace complaint
- Other consumer complaint requiring a custom route
-
Build the intake:
- Consumer name or preferred signature
- Merchant/company name
- Product or service
- Amount and currency
- Key dates: purchase, charge, cancellation, delivery, service failure, prior contact
- Order, invoice, account, ticket, or confirmation IDs
- Payment method and last four digits only, when relevant
- Desired resolution: refund, cancellation, replacement, correction, compensation, apology, account access
- Evidence list and prior support attempts
-
Audit evidence:
- Read
references/evidence-checklist.mdfor intake fields, evidence strength, and redaction rules. - Separate facts, assumptions, user claims, and missing documentation.
- Create a chronological timeline with dates where possible.
- Read
-
Choose an escalation route:
- Read
references/escalation-playbooks.mdwhen deciding the channel sequence. - Default sequence: merchant support -> supervisor/escalation team -> payment dispute or platform dispute -> regulator/BBB/state AG complaint draft -> attorney/small-claims referral if the user asks and the amount justifies it.
- For deadlines, formal filing requirements, or state/country-specific rules, verify current official sources before advising.
- Read
-
Draft the case pack:
- Case summary
- Evidence table
- Merchant/support message in the requested tone
- Payment dispute statement, if payment route is relevant
- Regulator/BBB/state AG complaint draft, if escalation is relevant
- Follow-up plan with dates and decision points
-
Review safety:
- Read
references/safety-boundaries.mdbefore drafting legal-adjacent, regulator-facing, financial, medical, identity-theft, or high-conflict materials. - Remove overclaims, legal conclusions, and unverifiable allegations.
- Read
Outputs
Produce concise, copy-ready materials. Prefer this order:
Case Summary: one page, neutral and factual.Evidence Gaps: what is missing and why it matters.Recommended Route: next best channel and fallback channel.Draft Message: one message, unless the user asks for multiple tones.Action Checklist: steps the user can approve.User Approval Needed: any action that would leave the chat or affect an account.
For tone, use:
friendly: cooperative, short, assumes good faith.firm: factual, cites evidence, gives a response deadline.final: last attempt before escalation, still professional and non-threatening.
Bundled Resources
references/evidence-checklist.md: Read when extracting facts, building a timeline, checking evidence quality, or deciding what to ask the user for.references/escalation-playbooks.md: Read when choosing merchant, payment, platform, BBB, regulator, or state attorney general routes.references/safety-boundaries.md: Read for legal-adjacent, high-stakes, fraud, identity, financial, medical, or hostile cases.assets/templates/: Copy-ready Markdown templates for case summaries, merchant requests, payment disputes, complaints, and follow-up logs.scripts/build_case_pack.py: Run when the user provides structured details or when you can create a JSON intake and the user wants files generated.
Script Usage
Use the script only after you have enough structured facts to avoid filling a pack with guesses.
python3 scripts/build_case_pack.py intake.json --out dispute-pack --tone firm
To show the expected input shape:
python3 scripts/build_case_pack.py --print-schema
For reproducible demos or tests, pass --today YYYY-MM-DD to fix generated follow-up deadlines.
Confirmation Gate
Before external action, present:
- Exact message/form content
- Recipient/channel
- Attachments or evidence to include
- Known risks or uncertainty
- A yes/no request for user approval
What ships with it: 10 files
21.9 KB alongside SKILL.md, 1 of them executable
agents/
- openai.yaml231 B
assets/
references/
- escalation-playbooks.md4.8 KB
- evidence-checklist.md3.5 KB
- safety-boundaries.md2.2 KB
scripts/
- build_case_pack.pyruns8.9 KB