maintain-tasks_skill

Overview

This skill provides task management for session continuity and coordination across multi-step work. It preserves working memory through compaction, maintaining intent, dependencies, status, and explicit subagent assignments. Use it to sequence work, assign subagents, and recover execution state after context summarization.

How this skill works

The skill stores Tasks as the session-level working memory that survives compaction: subjects, descriptions, dependencies, statuses, and subagent prefixes remain intact. Tasks are created for multi-step work, updated as status changes, and queried to determine what is actionable versus blocked. Explicit agent prefixes in task subjects ensure assignment and intent remain recoverable after context compaction.

When to use it

• Coordinating multi-step or multi-agent workflows within a session
• Preserving intent and state across context compaction or summary
• Assigning or tracking subagent responsibilities explicitly
• Sequencing work with dependencies (blockedBy / blocks)
• Managing tasks that will later be synchronized to external trackers

Best practices

• Always prefix the subject with an explicit subagent, e.g. "[engineer] Implement X"
• Create tasks for work that is 3+ steps or involves multiple handoffs
• Set status to in_progress before starting and mark completed immediately on finish
• Use blockedBy and blocks to encode sequence and unblock logic
• Include clear context and acceptance criteria in the description

Example use cases

• Sequential handoff: analyst researches → engineer implements → reviewer audits → tester validates
• Parallel feature work: multiple engineers implement components, tester waits (blockedBy both)
• Background reviews: assign a reviewer as a background agent with a task ID for retrieval
• Resumable work: include agent or resume IDs so an agent can continue after compaction
• Sync pattern: pull issues into Tasks, execute within session, then update external tracker at completion