Task Planner

Task Planner

Fetch the user's open Jira tickets and generate ready-to-paste prompts for kicking off work in Claude Code or Cursor sessions.

Step 1: Fetch Open Tickets

Use the mcp__mcp-atlassian__jira_search tool to find all tickets assigned to the current user that are not done:

JQL: assignee = currentUser() AND status != Done ORDER BY priority DESC, updated DESC
fields: summary,status,description,priority,issuetype,labels,project,parent
limit: 50

If the search returns many results, paginate to get them all.

Step 2: Present the Ticket Overview

Show a summary table of all tickets so the user can see their full backlog at a glance:

Group tickets by status (In Progress first, then To Do / Backlog / etc).

Step 3: Let the User Pick Tickets

Ask the user which tickets they want to generate prompts for. Options:

• All — generate prompts for every ticket
• By status — e.g., "just the In Progress ones"
• By selection — e.g., "PROJ-123 and PROJ-456"

Step 4: Fetch Full Details for Selected Tickets

For each selected ticket, use mcp__mcp-atlassian__jira_get_issue to get the full issue details including:

• Full description
• Acceptance criteria (often in description or a custom field)
• Comments (may contain implementation notes)
• Epic/parent context
• Labels

Step 5: Generate Copy-Paste Prompts

For each selected ticket, generate a self-contained prompt block inside a fenced code block that the user can copy and paste into a fresh Claude Code or Cursor session.

The prompt should follow this template:

```
[TICKET-KEY] Summary of the ticket

## Context
Brief description of what this ticket is about, drawn from the Jira description.
If this ticket has a parent epic, describe the epic's goal here for broader context.

## Acceptance Criteria
- Criteria extracted from the ticket description or comments
- If none exist, note that the user should define them

## Requirements
Detailed requirements from the ticket description, reformatted for clarity.

## Getting Started
1. Read the project's CLAUDE.md (or AGENTS.md) to understand conventions and structure
2. Start by running /grill-me to ask clarifying questions about ambiguous requirements
   before writing any code — resolve unknowns first
3. Create a new worktree branch off main for this work:
   `claude --worktree` or `git worktree add ../TICKET-KEY -b TICKET-KEY origin/main`
4. Explore the relevant codebase area before making changes
5. Implement the changes described above
6. Run the project's feedback commands (typecheck, lint, test) before committing
7. Create a commit with message format: [TICKET-KEY] type: description

## Jira
Ticket: TICKET-KEY
Link: https://forhims.atlassian.net/browse/TICKET-KEY
```

Prompt Generation Rules

• Extract structure from chaos: Jira descriptions are often messy. Parse them into clean sections (Context, Acceptance Criteria, Requirements).
• Include parent/epic context: If the ticket has a parent epic, mention it so the agent understands the bigger picture.
• Preserve technical details: Don't lose specific technical requirements, API contracts, file paths, or implementation hints from the description or comments.
• Add the Jira link: Always include the browseable link so the agent can reference it.
• Keep it self-contained: The prompt should work in a fresh session with no prior context — in Claude Code, Cursor, or any AI coding tool.
• Don't over-prescribe implementation: Give the "what" not the "how" — let the agent figure out the best approach.
• Always suggest /grill-me first: Every prompt should tell the agent to run /grill-me (or equivalent clarification step in Cursor) before diving into code, especially when acceptance criteria are vague or missing.
• Always suggest worktrees: Every prompt should instruct the agent to create a new branch off main in a worktree so work is isolated and doesn't conflict with other parallel sessions.
• Flag ambiguity: If the ticket description is vague or missing acceptance criteria, explicitly call it out and make /grill-me the mandatory first step.

Step 6: Offer Batch Output

After generating prompts, offer to:

• Print all prompts sequentially so the user can copy them one by one
• Write to files — save each prompt to a file like ~/prompts/TICKET-KEY.md for easy access later

Gotchas

• Some tickets may have empty descriptions — flag these and suggest the user flesh them out before kicking off work.
• Subtasks should reference their parent ticket for context.
• If a ticket is "In Progress", note that work may already be started — the prompt should mention checking for existing branches with git branch -a | grep TICKET-KEY.
• Respect the user's PR conventions from their CLAUDE.md (e.g., PR title format, Jira link in PR body).
• Cursor compatibility: The prompts work in any AI coding tool. For Cursor, /grill-me won't be available as a slash command — instead phrase it as "ask clarifying questions about ambiguous requirements before starting implementation".