#Intents and Canonical Identity

The AI produces a StructuredIntent, which is the normalized representation of a task.

#StructuredIntent Fields

• type: bug_report | feature_request | bug_fixed | unknown
• title: short, technical, canonical title
• description: short summary
• steps_to_reproduce: optional steps list
• tags: keywords for filtering
• canonical.object: the core object being affected
• canonical.action: the core action or failure
• resolution: only used for bug_fixed

#Canonical Identity

Canonical identity is used for de-duplication. If two intents share the same canonical.object and canonical.action, they are treated as the same task.

#Example

Input:

App crashes when opening monthly report
Bug: export button does nothing when list has 100+ items

Possible canonical identities:

• Task 1: object = "report", action = "open_crash"
• Task 2: object = "export_button", action = "no_response"

#Choosing analyze vs analyzeFlexible

• Use analyze() when you know the input is one task.
• Use analyzeFlexible() for real-world inputs where the model should decide.

#Related Docs

• AI Overview
• Usage