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
  

3. Step-by-Step Workflow

Follow these steps when helping a user build or update autolabeling rules:

Step 1: Ingest Requirements

• Ask the user what conversation properties or behaviors they want to classify (e.g., specific sub-agents, customer sentiment, authentication status, duration thresholds).
• Determine target GCP project ID and location.

Step 2: Draft or Edit Declarative YAML

• Create or update autolabel_rules.yaml in the user's workspace.
• Translate business logic into clear, prioritized CEL expressions.
• Ensure every rule ends with a fallback (condition: "").

Step 3: Compare with Active Remote Rules (diff)

Run diff to preview additions, modifications, and deletions:

uv run cxas insights diff-autolabel-rules --file autolabel_rules.yaml

Step 4: Dry-Run Deploy

Run push with --dry-run to verify API compatibility:

uv run cxas insights push-autolabel-rules --file autolabel_rules.yaml --dry-run

Step 5: Deploy to GCP Project (push)

Deploy the changes:

uv run cxas insights push-autolabel-rules --file autolabel_rules.yaml

(To delete remote rules that are no longer in the local YAML, append --force.)

Step 6: Export Existing Rules (pull)

To export existing active rules from an environment into YAML:

uv run cxas insights pull-autolabel-rules --parent projects/<PROJECT>/locations/<LOCATION> --out autolabel_rules.yaml

4. References & Tooling

• CEL Cookbook: Curated expressions and helper function recipes.
• JSON Schema: Schema definition for YAML linting.
• Design Document: Architectural specification.