todo

Todo List

Use TODO.md in the workspace root to track tasks that span multiple turns or involve sequential steps. TODO.md is injected into your system prompt every turn, so plans written here survive context truncation.

When to use

• Multi-step tasks (3+ steps, multiple tool calls)
• Sub-agent delegation (track what was spawned and what to do with results)
• Sequential work where order matters
• Cross-turn tasks that may take multiple conversation turns
• Anything you need to remember across context boundaries

Do NOT use for one-off questions or single-step tasks.

File format

## Task Title

- [ ] First step
- [ ] Second step (spawned: sub-agent "label")
- [x] Completed step

## Another Task

- [ ] Independent item

Each task is a ## heading with - [ ] checkbox items underneath.

Workflow

1. Before complex work: Read TODO.md, add a ## heading with broken-down steps.
2. Completing a step: Change - [ ] to - [x].
3. Finishing a task: When all items are checked, remove the entire section.
4. Checking status: Read TODO.md to see what's pending.

Sub-agent tracking pattern

When spawning sub-agents, anchor each one to a TODO item:

## Analyze projects

- [ ] canvas-it (spawned: "Analyze canvas-it") → store summary in KB
- [ ] contribart (spawned: "Analyze contribart") → store summary in KB
- [x] crush-tabs — done, stored in KB

This pattern ensures:

• You know which sub-agents are outstanding
• You know what to do when results arrive (the "→ action" note)
• Instructions survive context truncation because TODO.md is always re-read

Planning before acting

If a task involves creating new resources (skills, KB entries, files):

1. Search first — check if something similar already exists (kb search, list_dir)
2. Plan in TODO — write the steps before executing
3. Execute — work through items, checking them off
4. Clean up — remove the TODO section when done

Keep the file clean — remove completed task sections promptly.