document-redact

name	document-redact
description	Strip PII and sensitive entities from a document and store a redacted derivative.
inputs	[{"name":"document_id","type":"string","required":true,"description":"UUID of the Document to redact."},{"name":"entity_types","type":"array","required":false,"description":"Entity type filters to redact (e.g. [\"PERSON\", \"EMAIL\", \"SSN\"]). Defaults to all PII types."},{"name":"output_format","type":"string","required":false,"description":"Format of the redacted derivative — \"pdf\" or \"docx\". Defaults to \"pdf\"."}]
outputs	[{"name":"redacted_document_id","type":"string","description":"UUID of the newly created redacted DocumentInstance."},{"name":"redaction_count","type":"integer","description":"Number of entity spans replaced with redaction markers."}]
tools	["llm-redact","db-query","file-render"]

document-redact

When to use

Use this skill when the user needs a shareable version of a document with personal or confidential information removed. Always run document-classify first so the sensitivity tier is known; this skill reads that tier to determine which entity types to target by default.

How it works

• Fetch Page rows and their text content via db-query.
• Pass text in batches to llm-redact with the requested entity_types; receive spans to replace.
• Apply replacements and render the redacted output to file via file-render.
• Write a new DocumentInstance row pointing to the redacted file, linked to the source Document.
• Store a redaction manifest (count, entity types, page numbers) in the instance metadata.

Audit

Emits one ExecutionHistory row with event_type=DOCUMENT_EDITED and actor_type=agent after the redacted DocumentInstance row is committed.