task-management

Task Management

At the user level, we use Linear to manage tasks. When the Linear MCP is available, you can consult this directly.

Our workflow is to run many worktrees and checkouts of the repo and work on multiple research tasks and tasks in parallel. For managing todos and progress WITHIN a session you are encouraged to use bd (beads) to manage the work - falling back to a NOTES.md text based flow otherwise.

If available, use bd to manage subtasks and local todo lists (bd quickstart and bd --help)

• Break down plans into issues with dependencies and keep up-to-date during development
• Instruct subagents to consult bd regularly
• Use this to offload detail from your working memory and record extra details, bugs along the way with the user

The goal is to avoid 'clinging' to a context window, making it trivial to clear the session and re-hydrate context from bd. When you start a task, update FOCUS.md with a high-level understanding of the task. If a user, or another agent, reads your bd issues, commits and FOCUS.md they should be able to immediately take over and help with the work.