sensitive-redaction

name	sensitive-redaction
description	Detect and redact sensitive information from existing text or already parsed document content before sharing, archiving, model training, logging, ticket handoff, contract sharing, resume sharing, customer chat export, or external delivery. Use when the user asks for 脱敏, 敏感信息隐藏, 隐私清洗, 去标识化, 匿名化, 手机号脱敏, 邮箱脱敏, 身份证脱敏, 银行卡脱敏, 地址脱敏, 姓名脱敏, 公司名脱敏, 客户名脱敏, 订单号脱敏, 合同号脱敏, 日志脱敏, Token 脱敏, 密钥脱敏, password redaction, PII redaction, anonymization, de-identification, privacy cleanup, redact sensitive data, or mask secrets.
when_to_use	Use this skill only to redact sensitive information from text already available in the conversation, pasted by the user, or supplied by the platform attachment parser. Do not directly read, parse, OCR, or inspect uploaded files or images. If the user asks to export the redacted result as Word, PDF, Markdown, TXT, CSV, or another file, redact the content first and then route the redacted content to file-generator.
provider_type	builtin
provider_id	sensitive_redaction
runtime_type	hybrid
tools	["redact_text"]
max_calls_per_turn	5
timeout_seconds	10
display	{"icon":"shield-alert","category":"security_compliance","scenarios":["legal_compliance","document_handling","business_operations","customer_service"],"label":{"en_US":"Sensitive Redaction","zh_Hans":"敏感信息脱敏"},"description":{"en_US":"Designed for content that must be shared, archived, trained on, or exported safely; detects and redacts personal data, business identifiers, credentials, and other sensitive text.","zh_Hans":"适用于文本分享、归档、训练或导出前的安全处理，可识别并脱敏个人信息、业务标识、账号密钥和其他敏感内容。"},"when_to_use":{"en_US":"Use when existing text or parsed document content needs PII, business identifier, or secret redaction.","zh_Hans":"当已有文本或解析后的文档内容需要个人信息、业务标识或密钥脱敏时使用。"},"tags":{"en_US":["Privacy","Security","Redaction"],"zh_Hans":["隐私","安全","脱敏"]}}

Sensitive Redaction Skill

Use this skill to redact sensitive information from existing text. This skill does not read files, parse file formats, OCR images, generate files directly, or store original sensitive data.

Scope

Use this skill for:

• Customer chat logs, support tickets, call notes, and service records before external sharing.
• Contracts, resumes, business records, and parsed document content before forwarding or archiving.
• Application logs, API traces, error reports, and debugging snippets before issue handoff.
• Business data cleanup before model training, evaluation, demos, or public examples.
• Text that contains phones, emails, ID cards, bank cards, names, companies, customers, orders, contracts, addresses, secrets, tokens, passwords, IPs, or sensitive URL parameters.

Do not use this skill to:

• Directly read, parse, OCR, or inspect uploaded PDF, Word, spreadsheet, image, screenshot, or other file bytes.
• Preserve the original exact file layout or edit binary documents in place.
• Guarantee 100% detection of every sensitive field.
• Store or repeat complete original sensitive values.
• Generate a downloadable file directly.

Routing Rules

• If the user pasted text or the platform already supplied parsed document content, use this skill directly.
• If the user uploads a document and asks for redaction, use this skill only after the platform attachment parser has supplied source text in the conversation.
• If the user uploads an image or screenshot and asks for redaction, do not use this skill unless OCR or parsed text is already available.
• If the user asks to export the redacted result, first call redact_text, then pass only the redacted content to file-generator.
• If the user asks to summarize, classify, or transform the content after redaction, complete redaction first, then route the redacted content to the appropriate downstream skill.

Workflow

1. Confirm source text is available in the conversation or parsed attachment content.
2. Identify the redaction scenario and read exactly one reference before calling redact_text.
3. Choose a redaction level:
   • Use high for external sharing, public examples, training data, logs, contracts, resumes, customer data, legal/financial/HR data, or when unsure.
   • Use medium for internal sharing where partial traceability is acceptable.
   • Use low only when the user explicitly asks to preserve more context.