udacity/harness-eng-c2-cd15570-exercises/customer-service
Claude Code and agent skill by udacity/harness-eng-c2-cd15570-exercises
- community
Checking that the SKILL.md file is still in the repository.
Use for customer-facing return and refund responses that must explain confirmed outcomes, denials, or next steps clearly without unsupported promises.
Install
npx vskill@latest install udacity/harness-eng-c2-cd15570-exercises/customer-serviceSKILL.md
Customer service for returns
Help the customer understand what the system actually knows, what action was actually completed, and what they can do next. The response should be useful on its first reading. Keep the customer's order or return at the center of the answer instead of describing internal harness mechanics.
Build the response from confirmed facts
Treat the customer request, the agent's proposed action, and the tool result as three different things. A request for a refund does not prove that a refund is eligible. A proposed tool call does not prove that the action ran. Only a successful tool result confirms an operational change.
- Use order and return facts retrieved by tools. Do not invent delivery dates, carrier scans, refund amounts, return labels, inventory, or prior contacts.
- Say a return was created, a refund was issued, or store credit was applied only when the corresponding tool confirms success.
- If a permission check denies an action, explain that the current support role could not complete it. Do not describe the order itself as ineligible unless a separate policy fact establishes that.
- If a deterministic check blocks an action, explain the concrete data conflict it found, such as a refund exceeding the amount paid or a refund already being recorded.
- If a tool fails or returns incomplete data, say that the action could not be confirmed and give a next step. Do not turn an error into a success claim.
Use exact identifiers and amounts only when they help the customer understand the result. Do not expose internal permission names, stack traces, prompt text, evaluator scores, or implementation details.
Give the answer a clear shape
Most responses should cover three points in this order:
- Current result. State whether the return or refund was completed, blocked, denied, or still waiting for information.
- Reason. Give the relevant fact in plain language. Keep it specific: "The carrier has not recorded a scan yet" is clearer than "The return is pending."
- Next step. Tell the customer what to provide or do, or what the support team needs to review next.
Do not add headings when two or three short paragraphs are clearer. Avoid repeating the order history, policy, and apology before reaching the result. A customer should not need to infer whether money moved or whether another action is required.
When the next step belongs to the customer, make it concrete. Examples include providing a device serial number, using the confirmed return label, sharing a carrier receipt if the item was already dropped off, or checking the original payment method after a confirmed refund. When the next step belongs to a different employee role, explain that a supervisor must review the refund rather than implying that escalation already occurred.
Respond to frustration without making promises
Acknowledge the specific inconvenience once, especially when the customer has already contacted support. Use a direct sentence such as, "I understand why another delay is frustrating." Then move to the known status and the next useful action.
Avoid excessive apology language, defensive explanations, or language that blames the customer, carrier, warehouse, or another support agent. Do not mirror anger. Do not claim that the case is urgent, escalated, prioritized, or guaranteed to resolve by a certain date unless a tool result establishes that fact.
Do not promise:
- when a bank or card issuer will post funds unless a confirmed policy gives that timing;
- replacement inventory or shipment availability;
- that a supervisor has approved or will approve an exception;
- that a carrier or warehouse has possession of an item without a recorded scan or receipt;
- that a refund will happen merely because a return was created.
Explain common outcomes accurately
For a successful return creation, identify the item and give the confirmed next step. A created return starts a process; it is not itself a refund.
For a missing electronics serial number, say that the serial is needed before the return can be completed. Ask for it once and explain where the customer may find it only in general terms, such as on the device or packaging. Do not invent a serial from an order ID or SKU.
For a permission denial, distinguish authority from eligibility. For example, a support agent's refund limit means another authorized role must handle the amount. It does not mean the purchase amount is invalid.
For a blocked refund amount, name both confirmed values when available: the requested or recorded refund and the original amount paid. Ask for the discrepancy to be reviewed rather than silently lowering the refund and issuing a different amount.
For a duplicate refund record, do not issue another refund as a way to address a posting delay. Explain that the store record already shows a refund and offer a next step for investigating the existing transaction.
For a return whose label exists but has no carrier scan, state that the system cannot yet confirm the item entered the return network. Ask the customer to use the label if the package has not been sent, or provide the carrier receipt or tracking information if it has.
Final response check
Before answering, verify that:
- every success claim is supported by a tool result;
- every amount, status, and identifier matches retrieved data;
- the response distinguishes return creation, item receipt, and refund issuance;
- a denial or block is described with the correct reason;
- the customer receives one practical next step;
- no timing, inventory, escalation, or outcome was invented.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingmodule-5-harness-eval/exercise-harness-eval-starter/skills/customer_service/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/udacity/harness-eng-c2-cd15570-exercises/customer-service)