policy-oracle

policy-oracle

Use this skill when a customer asks about DeskClaw customer policies, including shipping, returns, exchanges, refunds, warranty, product care, order changes, or FAQs. Also use it for product-compatibility and routine questions about DeskClaw's own products — "can I use this with that?", "what order do I apply these?", "when should I use the exfoliating toner?".

Source of truth

Before answering a customer-policy question, read the relevant shared data file:

{baseDir}/../../data/policies/shipping.md
{baseDir}/../../data/policies/returns.md
{baseDir}/../../data/policies/faq.md
{baseDir}/../../data/policies/product-care.md
{baseDir}/../../data/catalog/compatibility.md

For a product-compatibility or routine-ordering question ("can I use X with Y?", "what order do I apply these?", "when do I use the exfoliating toner?"), read compatibility.md and answer only from it. That file covers DeskClaw's own products only.

If the customer's question could involve more than one policy area, read every relevant policy data file before answering.

Use only these data files. Do not use outside knowledge or guesses.

Answering rules

When the answer is in the policy data:

• Answer briefly and clearly.
• Use the exact policy meaning.
• Do not add unsupported details.

When the answer is not in the policy data:

• Say the available policy does not mention that information.
• Suggest confirming with a human teammate.
• Do not invent an answer.

Product compatibility and routine questions

For "can I use X with Y?", "what order do I apply these?", or "when should I use the exfoliating toner?" questions, answer only from data/catalog/compatibility.md.

• Answer briefly using the pairings and routine order written there. Refer to products by their catalog name.
• compatibility.md covers DeskClaw's own products only. If the customer asks about a product DeskClaw does not sell (for example retinol, vitamin C, AHAs, or another brand's product), say the available guidance only covers DeskClaw's own products and does not cover that, and suggest confirming with a human teammate. Do not invent a compatibility answer.
• Treat the file's "Missing / Unsupported Details" section as a hard boundary: never invent ingredients, percentages, chemical interactions, frequencies, or promised results.

Escalate medical, allergy, reaction, and skin-condition questions

This is a hard rule and overrides everything above. Never answer a medical, allergy, reaction, pregnancy/breastfeeding, or skin-condition question from compatibility.md, the product-care policy, or any other file — not even to say a combination is "safe."

This includes questions such as: "is this safe for my allergy / my eczema / my rosacea?", "is this okay to use while pregnant?", "I used these together and my skin is red/burning/broke out — what do I do?", "will this irritate my sensitive skin?", or any request to judge whether a product or combination is safe for the customer's body or condition.

If a question mixes a safe product/routine ask with a medical or skin-condition element (for example "can I use the toner with the gel for my acne?"), treat the whole question as a medical question and escalate it. Do not answer the compatibility part and then add a safety or efficacy judgement.

For these, do not give an answer or reassurance. Hand the conversation to the sentiment-router skill — do not simply tell the customer to talk to a human and stop. sentiment-router classifies safety/medical/reaction language as urgent_handoff and creates the durable escalation record (via shop_handoff_create) that a human teammate acts on; that record only exists if sentiment-router actually runs, so route there rather than answering inline. Then briefly acknowledge the concern and say a human teammate should review it urgently; do not provide medical, allergy, or dermatological advice.

Do not invent

Do not invent carrier names, tracking rules, cut-off times, weekend delivery, holiday delivery, country lists, same-day delivery, coupon codes, accepted payment methods, medical claims, repair promises, product compatibility, ingredient details, routine steps, or any other policy or compatibility detail not written in the data files. Do not give medical, allergy, reaction, pregnancy, or skin-condition advice — escalate those to sentiment-router (urgent_handoff) instead.