Developer Handover Pack

Developer Handover Pack

Document Retention and Archiving

Reads each incoming document and assigns a document type from the defined taxonomy, matches it to the correct retention schedule entry, and extracts key metadata including matter number, responsible party, retention period, and calculated destruction date. Outputs a structured metadata record that is written directly to the file in SharePoint and logged to the retention register in Microsoft 365. This agent handles the most error-prone portion of the current manual process by enforcing a single, consistent ruleset at the exact moment of document ingestion, eliminating the variability introduced by different staff members interpreting documents under time pressure. Estimated build time: 16 hours. Complexity: Complex.

// Input
document_source: 'sharepoint' | 'dropbox' | 'docusign'
document_id: string          // native file ID from source system
document_url: string         // direct download or preview URL
received_at: ISO8601         // timestamp of upload or envelope completion
matter_reference: string     // extracted from filename or envelope metadata where available

// Processing
// Agent reads document content (text extraction via Microsoft 365 or OCR fallback)
// Matches content against embedded document taxonomy ruleset
// Resolves document_type -> retention_period from structured ruleset table
// Calculates destruction_date = received_at + retention_period_days

// Output
document_type: string        // e.g. 'NDA', 'Employment Contract', 'Court Correspondence'
taxonomy_code: string        // internal taxonomy identifier, e.g. 'LEGAL-NDA-001'
retention_period_days: int   // e.g. 2555 (7 years)
destruction_date: ISO8601    // calculated end of retention period
matter_reference: string     // confirmed or extracted matter number
responsible_party: string    // derived from matter reference or envelope signer
archive_folder_path: string  // resolved destination path in SharePoint or Dropbox
register_row_id: string      // row ID written to Microsoft 365 retention register
classification_confidence: float  // 0.0 to 1.0; below 0.80 triggers human review flag
ocr_applied: boolean         // true if document required OCR pre-processing

Runs on a daily schedule and sweeps the full retention register in Microsoft 365 to identify every document whose destruction_date is on or before today's date. For each match it composes a structured Slack notification addressed to the correct legal manager, including document name, type, matter reference, destruction date, and the number of days the hold period has been served. The notification contains an approval prompt. Human approval is mandatory before any destructive action occurs. This agent replaces the monthly manual audit and the manual notification step, reducing the window between a destruction date passing and the responsible manager being informed from up to 31 days to under 24 hours. Estimated build time: 12 hours. Complexity: Moderate.

// Input
run_date: ISO8601             // today's date at execution time (UTC)
register_query_filter: string // destruction_date <= run_date AND status = 'active'

// Processing
// Agent queries Microsoft 365 retention register via Lists API
// Returns all rows where destruction_date <= run_date and status != 'destroyed' and status != 'deferred'
// For each matched row, resolves responsible_legal_manager Slack user ID from matter_reference lookup
// Constructs Slack Block Kit message with document_name, document_type, matter_reference,
//   destruction_date, days_overdue, sharepoint_link, and two action buttons: Approve / Defer
// Sends one message per document to the resolved Slack channel or DM
// Writes alert_sent_at timestamp to register row to prevent duplicate alerts on subsequent runs

// Output (per flagged document)
document_id: string           // register row identifier
document_name: string
document_type: string
matter_reference: string
destruction_date: ISO8601
days_overdue: int             // run_date minus destruction_date in days
responsible_manager_slack_id: string
slack_message_ts: string      // Slack message timestamp for threading approval response
alert_sent_at: ISO8601        // written back to register row

// On approval (Slack interactive callback)
approval_action: 'approve' | 'defer'
approver_slack_id: string
approved_at: ISO8601
// If approve: register row status set to 'pending_destruction', destruction_confirmed_by written
// If defer: register row status set to 'deferred', deferred_until date optionally supplied
// Outcome logged to register immediately; no file is deleted by this agent

// ============================================================
// DOCUMENT RETENTION AND ARCHIVING: END-TO-END DATA FLOW
// ============================================================

// TRIGGER SOURCES (three parallel entry points)
sharepoint_upload_event -> {
  document_id,
  document_url,
  document_source: 'sharepoint',
  received_at: timestamp,
  uploader_email
}

dropbox_webhook_event -> {
  document_id,
  document_url,
  document_source: 'dropbox',
  received_at: timestamp,
  folder_path
}

docusign_envelope_completed_webhook -> {
  envelope_id,
  document_url: signed_document_download_url,
  document_source: 'docusign',
  received_at: envelope_completed_at,
  matter_reference: custom_field['matter_reference'] || null,
  signer_name,
  signer_email
}

// --- HANDOFF TO: Document Classification Agent ---

// STEP 1: Text extraction
document_content_raw = extract_text(document_url)
// If extract_text returns null or low_quality:
  ocr_result = microsoft365_document_intelligence(document_url)
  document_content_raw = ocr_result.text
  ocr_applied = true
// If ocr_result.confidence < 0.60: quarantine_file(); log_error('unreadable'); halt()

// STEP 2: Taxonomy classification
classification_result = taxonomy_ruleset.match(document_content_raw)
// classification_result fields:
//   document_type: string          e.g. 'NDA'
//   taxonomy_code: string          e.g. 'LEGAL-NDA-001'
//   retention_period_days: int     e.g. 2555
//   classification_confidence: float

// STEP 3: Confidence gate
if classification_confidence < 0.80:
  move_to_review_folder(sharepoint_review_path)
  slack_alert(legal_admin_slack_id, 'classification_review_required', document_id)
  log_register_row(status: 'pending_review')
  halt()

// STEP 4: Metadata resolution
destruction_date = received_at + retention_period_days
matter_reference = matter_reference || parse_filename(document_id)
responsible_party = matter_lookup_table[matter_reference].owner
archive_folder_path = folder_routing_table[taxonomy_code][matter_reference]

// STEP 5: Write metadata to SharePoint
sharepoint.set_metadata(document_id, {
  document_type,
  taxonomy_code,
  retention_period_days,
  destruction_date,
  matter_reference,
  responsible_party,
  classified_at: now()
})

// STEP 6: Move to archive folder
if document_source == 'dropbox':
  dropbox.move(document_id, archive_folder_path)
else:
  sharepoint.move(document_id, archive_folder_path)

// STEP 7: Write retention register row
register_row_id = microsoft365_list.create_row({
  document_id,
  document_name,
  document_type,
  taxonomy_code,
  matter_reference,
  responsible_party,
  archive_folder_path,
  received_at,
  destruction_date,
  status: 'active',
  classified_at: now(),
  alert_sent_at: null
})

// --- CLASSIFICATION AGENT OUTPUT COMPLETE ---
// Register row is now the single source of truth for this document's lifecycle.

// ============================================================
// DAILY SCHEDULED RUN: Retention Schedule Monitor
// ============================================================

// --- HANDOFF TO: Retention Schedule Monitor ---

// STEP 8: Query retention register
run_date = today() // UTC 07:00
due_documents = microsoft365_list.query({
  filter: 'destruction_date <= run_date AND status == active AND alert_sent_at == null'
})
// Pagination: fetch all pages; max 5000 rows per request (Microsoft 365 List API limit)

// STEP 9: Resolve Slack recipients
for each doc in due_documents:
  slack_user_id = manager_lookup_table[doc.matter_reference] || fallback_channel_id
  days_overdue = date_diff(run_date, doc.destruction_date)

  // STEP 10: Send Slack alert
  slack_message_ts = slack.post_block_kit_message(slack_user_id, {
    document_name: doc.document_name,
    document_type: doc.document_type,
    matter_reference: doc.matter_reference,
    destruction_date: doc.destruction_date,
    days_overdue,
    sharepoint_link: resolve_link(doc.archive_folder_path, doc.document_id),
    actions: ['Approve Destruction', 'Defer']
  })

  // STEP 11: Update register with alert sent timestamp
  microsoft365_list.update_row(doc.register_row_id, {
    alert_sent_at: now(),
    slack_message_ts
  })

// ============================================================
// INTERACTIVE CALLBACK: Legal Manager Response
// ============================================================

// --- HANDOFF FROM: Slack interactive callback ---

slack_callback_payload -> {
  action: 'approve' | 'defer',
  slack_user_id: approver_slack_id,
  message_ts: slack_message_ts,
  document_id,
  register_row_id,
  deferred_until: ISO8601 | null   // supplied only on defer action
}

// STEP 12: Write outcome to register
if action == 'approve':
  microsoft365_list.update_row(register_row_id, {
    status: 'pending_destruction',
    destruction_confirmed_by: approver_slack_id,
    approved_at: now()
  })
  slack.post_reply(slack_message_ts, 'Destruction approved. Please proceed and confirm completion.')

if action == 'defer':
  deferred_until = deferred_until || run_date + 30_days
  microsoft365_list.update_row(register_row_id, {
    status: 'deferred',
    deferred_until,
    destruction_date: deferred_until,
    alert_sent_at: null    // reset so monitor re-alerts on new date
    deferred_by: approver_slack_id
  })
  slack.post_reply(slack_message_ts, 'Destruction deferred to ' + deferred_until)

// NOTE: No file is deleted by the automation at any point.
// Physical destruction and final register close-out remain human actions.
// ============================================================