societyer-document-intake

name	societyer-document-intake
description	Use when extracting or transposing documents into Societyer, especially Paperless-ngx OCR, scanned minutes, filings, finance documents, grants, privacy/HR records, or reviewed JSON bundles for /app/imports.

Societyer Document Intake

Use this skill to help an AI decide what Societyer can accept from source documents and how to stage it safely.

Required Workflow

1. Read the relevant support references before creating a bundle:
   • references/group-routing.md for high-level document routing, source cues, and safety rules.
   • references/page-field-support.md for page-by-page fields and support levels.
   • references/transposition-catalog.json for machine-readable bundle keys, record kinds, target tables, and accepted payload fields.
2. Prefer the /app/imports review pipeline for document-derived data:
   • importSessions.createFromBundle stages reviewed JSON.
   • paperless.createMeetingMinutesImportSession stages meeting-minute OCR.
   • paperless.createDiscoveryImportSession stages source document candidates.
   • paperless.createTransposedImportSession stages section-native records.
3. Do not treat extraction as final data entry. Stage records as Pending, include confidence, source IDs, and review notes, then require a human to approve before applying.
4. Use ISO dates (YYYY-MM-DD or ISO datetime where the app expects one) and store money as integer cents.
5. Preserve provenance on every document-derived record with sourceExternalIds such as paperless:123, and include a source record when possible.

Native Import Record Kinds

Societyer's import session pipeline supports these record kinds:

• Organization/history: source, fact, event, boardTerm, motion, budget
• Meeting-specific: meetingMinutes, meetingAttendance, motionEvidence
• Document catalog: documentCandidate
• Section records: filing, deadline, publication, insurancePolicy, financialStatement, financialStatementImport, grant, recordsLocation, archiveAccession, boardRoleAssignment, boardRoleChange, signingAuthority, budgetSnapshot, treasurerReport, transactionCandidate, sourceEvidence, secretVaultItem, pipaTraining, employee, volunteer

If a page is not listed as a native import target, do not invent a target. Use documentCandidate, sourceEvidence, or a clearly marked manual follow-up task.

Privacy Gates

• Raw secrets, passwords, recovery keys, certificate bodies, SINs, payroll source documents, banking details, and screening results must not be pasted into public or general notes fields.
• For credential material, create only a secretVaultItem external reference. Do not place the secret value in secretEncrypted; humans should enter encrypted material through the app UI.
• For restricted records, use sensitivity: "restricted" and avoid raw OCR excerpts unless the reference catalog explicitly allows it.
• Publications and transparency records must be reviewed separately before making anything public.

Output Expectations

When asked what a document can transpose, answer with:

• the most likely document group;
• supported target record kinds;
• exact payload fields the app can accept;
• fields that are present in the document but unsupported today;
• privacy or confidence flags;
• a sample import-session JSON bundle only when the user asks for bundle output.