customer-feedback-callback

Customer Feedback Callback

Use this skill when an agent needs to turn "call the customer after their order and see how it went" into a safe, repeatable workflow: one outbound CALL-E call per order, a structured result instead of a raw transcript, and a downstream decision step that a human still approves.

This is a post-delivery feedback callback, not a reminder, a sales call, or a collections call. It assumes the order/service already completed and the business wants a short, low-pressure check-in call.

Setup

• A configured CALL-E client (API key) with permission to create calls and read call results.
• A publicly reachable webhook URL the host registers with CALL-E for call-result delivery; without one, poll calls.get(id) instead of relying on handle_webhook.
• Nothing to install for the skill itself — references/examples.md shows the exact calls.create(...) call shape and a fake-client dry-run path with no live dependency.

When To Use

Use this skill for:

• one feedback call per completed order or service visit, dispatched automatically (for example 30 minutes after delivery)
• extracting a structured result (satisfaction score, sentiment, missing items, free-text summary) from the call instead of hand-parsing a transcript downstream
• distinguishing "customer did not pick up" from "customer answered and had nothing wrong" so only real signal reaches a human
• feeding call results into a small agent pipeline that cross-checks a new complaint against order history and recent trend before flagging it as urgent
• keeping the final "notify the business now" decision as a recommendation, never an automatic action

When Not To Use

Do not use this skill to:

• place the call before the order/service is actually complete
• retry a call that already reached the customer (completed, voicemail) — only a verified, terminal no_answer is retry-eligible, once; never retry on failed or any ambiguous/unresolved outcome — hold those for reconciliation instead (see Core Workflow)
• call about anything other than the specific completed order named in the task (no upsell, no collections, no marketing)
• auto-apply a discount, refund, or other compensation based on the call result — that always stays a separate, explicitly human-approved action
• store or forward the raw transcript to a human channel; only the structured result and a short generated summary should leave this workflow