Data Analytics

Protect personal data and secrets, constrain analytics tools, check output quality, require review and retain evidence.

Industry starter · Regulated SaaS, Private cloud

What it checks

Policies in the Data Analytics template, in the order the gateway runs them
PolicyOnChecksAction
prompt-injectionRequestDetects jailbreak and prompt-injection signalsBlock
pii-detectorRequest2 custom identifier patternsRedact
dlp-filterRequest3 blocked terms such as “private key” and “secret”Block
agent-firewallBothAllows 1 tool; denies 3 toolsBlock
tool-securityRequestBlocks unsafe tool definitions in the requestBlock
quality-scorerResponseAccuracy ≥ 0.80; relevancy ≥ 0.78; combined score ≥ 0.78; at least 120 characters and 2 sentencesBlock
human-oversightResponseNo action is set, so it does not review responsesNone
audit-loggerRequestAdds an allow-only audit marker to the decisionRecord
Policy YAML
pack:
  id: "cfg_data_analytics"
  name: "data-analytics"
  version: "0.1.0"
  enabled: true
  author: "verdictan"
  description: "Analytics controls for personal data, secrets, agent tools, output quality, and traceability."
policies:
  chain:
    - "prompt-injection"
    - "pii-detector"
    - "dlp-filter"
    - "agent-firewall"
    - "tool-security"
    - "quality-scorer"
    - "human-oversight"
    - "audit-logger"
policy:
  prompt-injection: {}
  pii-detector:
    action: redact
    detect_patterns:
      - "(?i)(?:api[_\\s-]?key|apikey)\\s*[:#=]?\\s*[A-Za-z0-9_\\-]{16,128}"
      - "(?i)(?:session[_\\s-]?id|sid)\\s*[:#=]?\\s*[A-Za-z0-9\\-]{8,128}"
  dlp-filter:
    blocked_terms:
      - "private key"
      - "secret"
      - "token"
    action: block
  agent-firewall:
    allowed_tools:
      - "document_search"
    blocked_tools:
      - "shell"
      - "code_exec"
      - "database_write"
  tool-security:
    analysis_mode: local
  quality-scorer:
    min_output_chars: 120
    min_sentences: 2
    thresholds:
      min_aggregate: 0.78
      min_relevancy: 0.78
      min_accuracy: 0.80
    weights:
      relevancy: 0.5
      accuracy: 0.5
  human-oversight: {}
  audit-logger: {}