Jira AI

Overview

The Jira AI system covers three operations:

Operation	Endpoint	Description
Generate	`POST /api/canvas/jira/generate`	Step-by-step structured work item generation
Analyze	`POST /api/canvas/jira/analyze`	Inline AI suggestions for a single work item
Rewrite	`POST /api/canvas/jira/rewrite`	Field-level rewrite of selected text in an item

Generation

Jira generation produces structured output validated against a Zod schema:

{
  "items": [
    {
      "key": "EPIC-001",
      "type": "epic",
      "summary": "Reduce checkout friction",
      "description": "...",
      "acceptanceCriteria": ["..."],
      "noGos": ["..."],
      "sourceNoteLinks": [{ "noteId": "...", "noteTitle": "..." }]
    }
  ],
  "recommendedNextStep": "stories"
}

Generated items are persisted via the standard sync pipeline (same D1 push as notes).

Generation modes

Mode	Body field	What it does
`generate_step`	`step`	Generates the next tier of the hierarchy
`regenerate_items`	`itemIds`	Improves selected existing items
`generate_children`	`parentId`	Decomposes a parent item into children

Recommended generation sequence

Ideas — high-level opportunity areas
Discoveries — research findings and hypotheses
Prototypes — experiment designs
Epics and Stories — user-facing outcomes and requirements
Tasks and Subtasks — implementation work items

The "Generate next step" button in the Jira Action Bar automatically identifies which tier is missing and suggests the right generation.

Analyze

The analyze endpoint produces inline suggestions for a work item section:

{
  "suggestions": [
    {
      "sectionText": "Users can reset their password",
      "advice": "Add a Given/When/Then format for this acceptance criterion",
      "actions": [
        { "label": "Add Given/When/Then", "prompt": "Rewrite this criterion with Given/When/Then" }
      ]
    }
  ]
}

Suggestions are rendered as @subagent annotation blockquotes inside the work item editor. Each suggestion has action buttons that trigger a rewrite call.

Rewrite

The rewrite endpoint accepts a selected piece of text and a free-form instruction, and returns a rewritten version:

{ "rewritten": "Given a user on the login page, when they click Forgot password..." }

The client replaces the selected text and highlights it yellow to indicate an AI-applied change.

Quota and billing

All Jira AI endpoints consume AI credits. Charges are applied to the board owner's monthly credit quota, not the requesting user. This means collaborators can use AI without needing a Pro plan themselves.