linear-create-issue

name: linear-create-issue description: Create Linear issues intelligently - search for duplicates, learn team patterns, evaluate sub-issue breakdown, then create with proper context. target: linear_agent

Linear Create Issue

When to Use

• User asks to "create an issue" or "file a bug"
• User asks to "add a ticket" or "track this"
• User describes a problem, task, or feature to be tracked
• User wants to create issues with sub-tasks

Tools

Discovery & Context

• LINEAR_CUSTOM_GET_WORKSPACE_CONTEXT — Get teams, projects, labels, states
• LINEAR_CUSTOM_RESOLVE_CONTEXT — Map fuzzy names → IDs (team, user, labels, project, state)
• LINEAR_CUSTOM_SEARCH_ISSUES — Search issues by keyword
• LINEAR_CUSTOM_GET_ISSUE_FULL_CONTEXT — Get full issue details
• LINEAR_CUSTOM_GET_ACTIVE_SPRINT — Get current sprint/cycle info
• LINEAR_CUSTOM_GET_MY_TASKS — Get authenticated user's assigned issues

Creation

• LINEAR_CUSTOM_CREATE_ISSUE — Create issue with all fields
  • Required: team_id, title
  • Optional: description, assignee_id, priority (0-4), state_id, label_ids, project_id, cycle_id, due_date, estimate, parent_id
  • Sub-issues: sub_issues array [{title, description, assignee_id, priority}]

Workflow

Step 1: Search for Duplicates (MANDATORY)

Before creating anything, search for existing issues:

LINEAR_CUSTOM_SEARCH_ISSUES(query="<keywords from user request>")

If similar issues exist:

• Show them to the user: "I found these existing issues that look similar..."
• Ask if they want to proceed or update an existing one
• Only continue to creation after user confirms no duplicate

Step 2: Learn Team Patterns

Understand how the team structures issues:

1. Get workspace context:

   LINEAR_CUSTOM_GET_WORKSPACE_CONTEXT()
   

   This reveals team names, available states, labels, and projects.

2. Read recent team issues (2-3):

   LINEAR_CUSTOM_SEARCH_ISSUES(query="recent")
   

   Study the results to learn:

   • Title format (e.g., "[Component] Description" vs plain descriptions)
   • Description structure (bullet points? acceptance criteria? steps to reproduce?)
   • Common labels used
   • Priority conventions

3. Match the team's style when writing the title and description.

Step 3: Evaluate Sub-Issue Potential

If the user's request is broad or multi-faceted:

• Break it down: Suggest logical sub-issues
  • Example: "Implement auth" → "Design login flow", "Implement OAuth", "Add MFA", "Write tests"
• Ask the user: "This looks like it could be broken into sub-tasks. Want me to create sub-issues?"
• Keep it practical: Only suggest 2-5 sub-issues; too many is unhelpful

If the request is specific and focused, skip sub-issues entirely.

Step 4: Resolve All Entities

Use RESOLVE_CONTEXT to convert names to IDs:

LINEAR_CUSTOM_RESOLVE_CONTEXT(
    team_name="engineering",
    user_name="john",
    label_names=["bug", "critical"],
    project_name="Q1 Sprint",
    state_name="backlog"
)

This returns: team_id, user_id, label_ids, project_id, state_id

For cycle_id (adding to current sprint):

LINEAR_CUSTOM_GET_ACTIVE_SPRINT()

Step 5: Create the Issue

LINEAR_CUSTOM_CREATE_ISSUE(
    team_id=resolved_team_id,
    title="Formatted like team's convention",
    description="Structured like team's existing issues",
    assignee_id=resolved_user_id,
    priority=2,
    label_ids=resolved_label_ids,
    project_id=resolved_project_id,
    state_id=resolved_state_id,
    cycle_id=sprint_cycle_id,
    sub_issues=[
        {"title": "Sub-task 1", "priority": 2},
        {"title": "Sub-task 2", "priority": 2}
    ]
)

Step 6: Confirm Results

Report back clearly:

• Issue identifier (e.g., "ENG-456")
• URL to the issue
• What was set (assignee, labels, project, sprint)
• Sub-issues created (if any)
• Anything that needs follow-up

Priority Mapping

• 0 = No priority
• 1 = Urgent
• 2 = High
• 3 = Medium
• 4 = Low

Important Rules

1. Always search first — Never create without checking for duplicates
2. Learn before creating — Read team patterns to match style
3. Never guess IDs — Always use RESOLVE_CONTEXT to convert names
4. Suggest structure — Offer sub-issues for broad tasks
5. Confirm results — Show what was created with identifiers and URLs