Developer Handover Pack

Developer Handover Pack

Regulatory Change Monitoring

Runs on a daily schedule to fetch new content from all configured regulatory feeds and websites via Feedly, deduplicates items already seen in the Airtable intake table, and passes only net-new content downstream to the Relevance and Summary Agent. This agent eliminates the need for any manual website visiting and is the entry point for the entire automation pipeline. It must be stable before any downstream agent build begins.

// Input
feedly.stream_id       : string   // Feedly collection ID for regulatory sources
feedly.newer_than      : unix_ms  // Timestamp of last successful poll (stored in Airtable config row)
feedly.count           : integer  // Max items per poll, default 50

// Deduplication lookup (Airtable)
airtable.table         : 'Regulatory_Intake'
airtable.filter        : {feedly_article_id} = '{article.id}'

// Output (passed to Relevance and Summary Agent)
article.id             : string   // Feedly article fingerprint
article.title          : string
article.source_name    : string   // Feed label as configured in Feedly collection
article.source_url     : string
article.published_at   : ISO8601
article.full_text      : string   // Extracted body text, max 4000 chars
article.canonical_url  : string
article.is_duplicate   : boolean  // true = skip; false = pass downstream

Receives each net-new article from the Source Monitor Agent and applies an AI classification step to score relevance against the business's configured jurisdiction profile and topic tag list. Items that score below the configured threshold are discarded silently. Items above the threshold receive a two-to-three sentence plain-language summary, an extracted or inferred effective date, and a suggested internal owner derived from the topic-to-owner routing table stored in Airtable. Borderline items (within a configurable band around the threshold) are held and routed to the Compliance Coordinator for a 5-minute manual review before any further action is taken.

// Input (from Source Monitor Agent)
article.id             : string
article.title          : string
article.source_name    : string
article.published_at   : ISO8601
article.full_text      : string
article.canonical_url  : string

// Classification prompt context (retrieved from Notion page 'Relevance Config')
config.jurisdictions   : string[] // e.g. ['AU', 'NZ', 'UK']
config.topic_tags      : string[] // e.g. ['AML', 'Privacy', 'Financial Services', 'OH&S']
config.keywords        : string[] // Boosted signal terms
config.threshold_pass  : float    // default 0.70
config.threshold_review: float    // default 0.50 (below this = discard)

// Routing table lookup (Airtable table: 'Routing_Rules')
routing.topic_tag      : string   // matched tag
routing.owner_name     : string
routing.owner_slack_id : string
routing.owner_email    : string

// Output (written to Airtable table: 'Regulatory_Register')
register.article_id    : string
register.title         : string
register.source_name   : string
register.source_url    : string
register.published_at  : ISO8601
register.effective_date: ISO8601 | null  // extracted by AI; null if not found
register.topic_tag     : string
register.relevance_score: float
register.summary       : string   // 2-3 sentence plain-language summary
register.suggested_owner: string
register.owner_slack_id: string
register.status        : enum('pending_review', 'approved', 'discarded')
register.flagged_borderline: boolean

// On borderline (threshold_review <= score < threshold_pass)
// Pause pipeline; send borderline alert to Compliance Coordinator Slack DM
// Await manual status update in Airtable (approved | discarded) before Alert Agent fires

Handles all outbound communication for the pipeline. It fires immediately after a new record with status = 'approved' is written to the Airtable Regulatory_Register, posting a structured Slack alert tagged to the responsible owner in the correct topic channel. On a Monday morning schedule it assembles the previous week's approved items into a digest email and sends it via Gmail to the configured management recipients. It also monitors the register for action items that remain unacknowledged beyond a configurable number of days and sends overdue follow-up Slack messages to the responsible owner.

// Input (real-time alert path)
register.record_id     : string
register.title         : string
register.summary       : string
register.effective_date: ISO8601 | null
register.topic_tag     : string
register.relevance_score: float
register.source_url    : string
register.owner_slack_id: string
register.suggested_owner: string

// Input (weekly digest path)
// Query Airtable Regulatory_Register WHERE created_at >= last Monday 00:00
// AND status = 'approved'
// ORDER BY relevance_score DESC

// Input (overdue follow-up path)
// Query Airtable Regulatory_Register WHERE acknowledged_at IS NULL
// AND status = 'approved'
// AND created_at < NOW() - overdue_days
config.overdue_days    : integer  // default 7; stored in Airtable config table

// Output (real-time alert)
slack.channel          : string   // resolved from topic_tag via Routing_Rules
slack.mention          : string   // owner_slack_id formatted as <@U...>
slack.blocks           : object[] // Slack Block Kit message with title, summary, effective date, source link, and acknowledge button

// Output (weekly digest)
gmail.to               : string[] // management recipient list from Airtable config
gmail.subject          : string   // 'Weekly Regulatory Digest – {ISO week label}'
gmail.html_body        : string   // Templated HTML listing each item: title, summary, owner, effective date, register link

// Output (overdue follow-up)
slack.dm_to            : string   // owner_slack_id
slack.message          : string   // Plain text overdue reminder with item title and register deep link

// On acknowledge (Slack interactive button callback)
airtable.update        : {acknowledged_at: ISO8601, acknowledged_by: slack_user_id}

// ============================================================
// REGULATORY CHANGE MONITORING: END-TO-END DATA FLOW
// ============================================================

// TRIGGER
// Scheduled daily at 07:00 local (configurable)
// OR Feedly webhook fires on new item (Pro+ plan required)
TRIGGER -> Source Monitor Agent
  inputs:
    feedly.stream_id        = 'user/{feedly_user_id}/category/Regulatory_Sources'
    feedly.newer_than       = airtable.config['last_polled_at']  // unix_ms
    feedly.count            = 50

// ============================================================
// AGENT HANDOFF 1: Source Monitor Agent -> Deduplication
// ============================================================
  for each article in feedly.response.items:
    article.id              -> airtable.lookup('Regulatory_Intake', feedly_article_id)
    IF match found:
      article.is_duplicate  = true  // skip; no downstream call
      log('duplicate_skipped', article.id)
    ELSE:
      airtable.insert('Regulatory_Intake', {
        feedly_article_id   : article.id,
        title               : article.title,
        source_name         : article.source_name,
        source_url          : article.canonical_url,
        published_at        : article.published_at,
        full_text           : article.full_text[0:4000],
        intake_status       : 'queued'
      })
      article.is_duplicate  = false
      -> pass to Relevance and Summary Agent

  airtable.config['last_polled_at'] = NOW()  // update poll timestamp

// ============================================================
// AGENT HANDOFF 2: Source Monitor Agent -> Relevance and Summary Agent
// ============================================================
Relevance and Summary Agent receives:
  article.id, article.title, article.source_name,
  article.published_at, article.full_text, article.canonical_url

  // Step A: Fetch classification config from Notion
  notion.page('Relevance Config') ->
    config.jurisdictions    : string[]
    config.topic_tags       : string[]
    config.keywords         : string[]
    config.threshold_pass   : float    // from Airtable config table
    config.threshold_review : float    // from Airtable config table

  // Step B: AI classification call
  prompt_input:
    system  : 'You are a regulatory compliance classifier. Score relevance 0.0-1.0.'
    context : config.jurisdictions, config.topic_tags, config.keywords
    content : article.title + article.full_text
  prompt_output:
    relevance_score         : float
    matched_topic_tag       : string
    reasoning               : string   // logged, not surfaced

  // Step C: Route on score
  IF relevance_score >= threshold_pass:
    classification_status   = 'approved'
  ELSE IF relevance_score >= threshold_review:
    classification_status   = 'borderline'  // hold for human review
  ELSE:
    classification_status   = 'discarded'
    log('item_discarded', article.id, relevance_score)
    STOP

  // Step D: Summary generation (approved + borderline only)
  prompt_input:
    system  : 'Write a 2-3 sentence plain-language summary of this regulatory change.'
    content : article.full_text
  prompt_output:
    summary                 : string
    effective_date          : ISO8601 | null

  // Step E: Routing table lookup
  airtable.lookup('Routing_Rules', {topic_tag: matched_topic_tag}) ->
    routing.owner_name      : string
    routing.owner_slack_id  : string
    routing.owner_email     : string
  FALLBACK: topic_tag = 'Unclassified' -> Compliance Coordinator

  // Step F: Write to Regulatory_Register
  airtable.insert('Regulatory_Register', {
    article_id              : article.id,
    title                   : article.title,
    source_name             : article.source_name,
    source_url              : article.canonical_url,
    published_at            : article.published_at,
    effective_date          : effective_date,
    topic_tag               : matched_topic_tag,
    relevance_score         : relevance_score,
    reasoning               : reasoning,
    summary                 : summary,
    suggested_owner         : routing.owner_name,
    owner_slack_id          : routing.owner_slack_id,
    owner_email             : routing.owner_email,
    status                  : classification_status,
    flagged_borderline      : (classification_status == 'borderline'),
    acknowledged_at         : null,
    acknowledged_by         : null,
    created_at              : NOW()
  })

  IF classification_status == 'borderline':
    slack.dm(owner_slack_id = COMPLIANCE_COORDINATOR_SLACK_ID,
             message = 'Borderline item requires review: {title} | {airtable_record_link}')
    PAUSE pipeline for this item; await manual status update in Airtable
  ELSE IF classification_status == 'approved':
    -> pass record_id to Alert and Digest Agent (immediate path)

// ============================================================
// AGENT HANDOFF 3: Relevance and Summary Agent -> Alert and Digest Agent
// ============================================================
Alert and Digest Agent: immediate path
  trigger          : Airtable record status updated to 'approved'
  inputs           : register.record_id -> fetch full record

  // Post Slack Block Kit alert
  slack.post({
    channel        : resolve_channel(matched_topic_tag),  // from Routing_Rules
    blocks: [
      header       : 'New Regulatory Change: {title}',
      section      : summary,
      fields       : [effective_date, source_name, topic_tag, relevance_score],
      actions      : [button('Acknowledge', value=record_id)]
    ]
  })

Alert and Digest Agent: weekly digest path
  trigger          : Monday 08:00 scheduled run
  query            : Airtable Regulatory_Register WHERE
                     created_at >= last_monday AND status = 'approved'
                     ORDER BY relevance_score DESC
  gmail.send({
    to             : airtable.config['digest_recipients'],
    subject        : 'Weekly Regulatory Digest – {ISO_week_label}',
    html_body      : render_digest_template(items)
  })

Alert and Digest Agent: overdue follow-up path
  trigger          : Daily 09:00 scheduled run
  query            : Airtable Regulatory_Register WHERE
                     acknowledged_at IS NULL AND status = 'approved'
                     AND created_at < NOW() - config.overdue_days
  for each record:
    IF last_overdue_sent_at < NOW() - 24h OR last_overdue_sent_at IS NULL:
      slack.dm(owner_slack_id, 'Overdue action: {title} | {airtable_record_link}')
      airtable.update(record_id, {last_overdue_sent_at: NOW()})

// ============================================================
// SLACK INTERACTIVE CALLBACK (acknowledge button)
// ============================================================
  slack.callback.payload.actions[0].value = record_id
  slack.callback.user.id                  = slack_user_id
  -> airtable.update('Regulatory_Register', record_id, {
       acknowledged_at  : NOW(),
       acknowledged_by  : slack_user_id
     })

// ============================================================
// ERROR LOGGING (all agents)
// ============================================================
  ON any agent error:
    airtable.insert('Error_Log', {
      agent_name       : string,
      error_type       : string,
      error_message    : string,
      article_id       : string | null,
      timestamp        : ISO8601
    })
    IF severity == 'critical':
      slack.dm(FULLSPEC_ALERT_CHANNEL, 'Automation error: {error_message}')