cxas-autolabel-rules

CCAI Insights Autolabeling Rules Skill

This skill guides you in authoring, refining, validating, and synchronizing Contact Center AI (CCAI) Insights Autolabeling Rules.

Autolabeling rules enrich ingested conversations with custom key-value metadata evaluated via Common Expression Language (CEL).

---

1. Overview & Declarative YAML Schema

Rules are defined declaratively in autolabel_rules.yaml:

version: "1.0"
project_id: "your-gcp-project-id"
location: "us-central1"

autolabeling_rules:
  - rule_id: "agent_domain"
    display_name: "Agent Domain Classifier"
    label_key: "agent_domain"
    label_key_type: "LABEL_KEY_TYPE_CUSTOM"
    active: true
    conditions:
      - condition: "containsSubAgent(conversation, 'billing_specialist')"
        value: "'billing'"
      - condition: "containsSubAgent(conversation, 'tech_support')"
        value: "'tech_support'"
      - condition: ""
        value: "'general'"

Core Schema Requirements

1. rule_id: Unique alphanumeric identifier (snake_case or kebab-case).
2. label_key: The metadata key name to attach to conversation.labels.
3. conditions: Ordered array of conditions evaluated top-to-bottom.
   • condition: CEL boolean expression (e.g. conversation.duration > 300).
   • value: CEL expression or quoted string literal (e.g. 'vip', 'escalated').
   • Mandatory Fallback: The final condition in every rule MUST be condition: "" to act as the default fallback value.

---

2. Common Expression Language (CEL) Authoring Rules

Refer to the CEL Cookbook for full syntax and function references.

Built-in Helper Functions

• Sub-Agent / Flow Detection:

  containsSubAgent(conversation, "billing_subagent")
  

• Session Parameter Matching:

  hasSessionParam(conversation, "authenticated", "true")
  

• Sentiment Analysis:

  hasCallerSentiment(conversation, "NEGATIVE")
  

• Turn & Duration Checks:

  conversation.duration > 300 && conversation.turnCount >= 10