敏感字段脱敏Redact sensitive fields

识别并脱敏敏感字段，保留可审计记录。 输入：原文、敏感字段清单、脱敏格式。

Output redacted text and a field log; do not guess unlisted fields. Inputs: Source text, sensitive-field list, and redaction format.

按清单 {fields} 脱敏原文。输出 redacted_text 与 redactions[]：field、location、replacement。只处理清单内字段；疑似但不在清单的字段列 review[]，不自行脱敏。replacement 使用 {format}，不得保留可逆原始值。

占位符用 {字段} 标注，运行前替换为本次输入。

Redact the source according to {fields}. Output redacted_text and redactions[] with field, location and replacement. Handle only listed fields; put suspected but unlisted fields in review[] without redacting them. replacement uses {format} and must not retain reversible original values.

Replace {field} placeholders with the current inputs before running.

redacted_text、redactions[]、review[]。

redacted_text, redactions[], and review[].

• 清单字段全部替换且不可逆。
• review 项未被擅自处理。
• location 可回连原文。

• Listed fields are replaced irreversibly.
• Review items are not handled without authorisation.
• Locations link back to the source.

• 漏脱敏：补字段清单重跑。
• 过度脱敏：把非清单项移回 review。

• Missed redaction: extend the field list and rerun.
• Over-redaction: move unlisted items back to review.

提示词效果未实测。机制依据见出处；换模型、版本或输入会改变结果。

This site has not tested the effect of this prompt. See the sources for the mechanism; changing model, version or input can change the result.