负反馈回复安全草稿

Use this skill after feedback triage when an operator needs a calm, evidence-bound draft and an escalation reminder.

Prerequisites

• - A redacted consumerfeedbacktriage.json.
• An owner policy containing approved tone, support channel, and allowed compensation wording.

Input and output

Input is a triage queue; only pseudonymous references and cause fields are copied into the output. No raw customer message is needed.

Output is feedbackreplyguard.json with a draft, evidence checklist, escalation flag, and fixed send_executed=false.

Procedure

1. 1. Run consumer-feedback-triage or load an equivalent redacted queue.
2. Run python scripts/buildfeedbackreplyguard.py --input triage.json --output feedbackreply_guard.json.
3. A human checks the draft against the current platform rules and sends or edits it in the approved channel if appropriate.

Error recovery

Malformed input or missing HUMANREVIEWREQUIRED status exits non-zero. Missing policy wording produces a neutral placeholder rather than a promise of refund, compensation, deletion, or liability.

Permission and privacy boundaries

This skill never sends a reply, contacts a consumer, issues a refund, changes a rating, hides a review, requests a positive review, or exposes private information. It does not decide whether a complaint is true.

Acceptance

The example must create a neutral draft for each queue item, preserve the escalation flag for HIGH or CRITICAL issues, and show send_executed=false.