Task Tracker Skill

Description

Automatic intelligent task tracking for Claude Code sessions. This skill silently manages a personal kanban board with context awareness, smart analysis, and agent integration.

Trigger

This skill is model-invoked. Analyze each user prompt and decide whether to create/update tasks.

Rules

When to CREATE a task:

• User gives an actionable instruction: "fix login bug", "add dark mode", "refactor the auth module"
• User describes work to be done: "we need to update the API", "the form validation is broken"

When NOT to create a task:

• User asks a question: "how does auth work?", "what does this function do?"
• User asks for explanation or review: "explain this code", "review this PR"
• User gives a meta-command: "/board", "/task done", any slash command
• User is chatting casually

Smart Task Creation:

Use task_smart_create instead of task_create. It automatically:

• Generates a clean title from messy prompts ("this is slow" → "Optimize performance of [module]")
• Finds related files in the project
• Auto-tags based on content (#bug, #feature, #performance, #security, #ui, etc.)
• Sets priority based on urgency keywords and code hotspots
• Saves initial analysis as task context

Deduplication:

Before creating a new task, check existing tasks using task_list. If a similar task already exists, update it instead.

Auto-status:

• When creating a task from an actionable prompt, set status to "doing" (user is about to work on it)
• Only 1 task should be "doing" at a time. If there's already a "doing" task:
  1. Save its context with task_context_save (what was done, what's left)
  2. Move it to "todo" with task_move
  3. Then create the new task

Context Management:

• When SWITCHING tasks (moving current "doing" to "todo"):
  • Call task_context_save on the old task with: files you examined, decisions made, remaining work
• When RESUMING a task (moving a "todo" back to "doing"):
  • Call task_context_load to get the full context back
  • Use the loaded context to continue work seamlessly

Task Review on Completion:

• When user says "done" or you complete the task's objective:
  • Call task_review to get a pre-close analysis
  • The review checks: files changed, commits made, TODOs left, tests exist, types check
  • If all checks pass, move to done silently
  • If there are warnings, mention them briefly before closing

Agent Creation:

• For complex tasks (multiple files, estimated >30 min), consider creating an agent:
  • Call task_agent_create with constraints and test commands
  • This creates a .claude/agents/task-*.md file that can be launched in any session
• When a task with an agent is completed:
  • Call task_agent_delete to clean up

SILENCE:

• Do NOT mention task tracking to the user unless they explicitly ask about it
• Do NOT show task IDs, board status, or any tracking info in responses
• The tracking is completely invisible to the user
• Only surface information when user uses /soloboard:board, /soloboard:task, or /soloboard:project
• Analysis results (related files, suggested approach) should inform YOUR work, not be shown to the user

Auto-initialization:

At the START of every session, before doing anything else:

1. Call auto_init — it safely creates .kanban dir, project board, and session (idempotent)
2. This is SILENT — never mention it to the user

Smart Task Creation Flow:

1. Call auto_init (if not called yet this session)
2. Call task_list to check for duplicates
3. If there's already a "doing" task: a. Call task_context_save on it b. Call task_move to move it to "todo"
4. Call task_smart_create with the user's prompt (NOT task_create)
5. Use the analysis from task_smart_create to inform your work approach
6. All SILENTLY — user sees nothing about tracking

Session Management:

• auto_init handles session creation automatically
• Hooks handle file tracking, commit linking, and session closing
• No manual session management needed