paulnsorensen/triage

Todoist Triage

Process overdue and stale tasks interactively. You present, the user decides — never assume a task's fate.

Flow

Step 1: Fetch Overdue Tasks

Spawn todoist-fetch to retrieve tasks in a fresh context window:

Agent(subagent_type: "todoist-fetch", prompt: "Fetch overdue tasks: find-tasks with filter 'overdue'. Sort by age (oldest first). Include task IDs, titles, projects, priorities, due dates, creation dates, descriptions, and any labels.")

If the user said "triage inbox" or "inbox zero":

Agent(subagent_type: "todoist-fetch", prompt: "Fetch inbox tasks: find-tasks with filter '#Inbox'. Sort by creation date (oldest first). Include task IDs, titles, priorities, due dates, creation dates, descriptions, and any labels.")

Step 2: Calculate Staleness

For each task, calculate days overdue. Flag tasks that are 3+ weeks overdue — these have been implicitly deprioritized and are strong candidates for deletion or Someday.

Step 3: Present Batches

Show 5 tasks at a time. For each task, display:

### Task 1/N: [Task Title]
- Project: [project name]
- Priority: [p1-p4]
- Due: [original due date] (X days overdue)
- Description: [if any, truncated to 2 lines]
- Staleness: [STALE if 3+ weeks — "This task has been sitting for X weeks. Honest question: is this still relevant?"]

Actions: (C)omplete | (R)eschedule [when?] | (D)elete | (S)omeday | (K)eep as-is | (?)Research

Use AskUserQuestion to get the user's decision. Accept shorthand: c, r tomorrow, r next monday, d, s, k, ?.

Step 4: Handle "Research" Requests

If the user picks ? on a task, invoke the research skill to get context:

Skill(skill="todoist-flow:research", args="Context for this task: '[task title]'. [task description if any]. Help the user decide if this is still relevant and what the current state of this topic is.")

After research returns, re-present the task with the research context and ask again.

Step 5: Execute Decisions

Batch decisions and run through the write pipeline (distill → scribe → QA):

Operation mapping:

• Complete: complete-tasks
• Reschedule: reschedule-tasks (NEVER update-tasks — preserves recurrence)
• Delete: delete-object
• Someday: Move to "Someday" project (create if needed) + strip due date via update-tasks with dueString: "remove"
• Keep: No action

1. Validate reasoning — spawn todoist-distill:

Agent(subagent_type: "todoist-distill", prompt: "Validate these triage decisions against user intent: [decisions with task data, user choices, and context]")

2. Format commands — spawn todoist-scribe with validated plan:

Agent(subagent_type: "todoist-scribe", prompt: "Format these validated triage operations as MCP commands: [distill's validated plan]")

3. Verify and execute — spawn todoist-qa:

Agent(subagent_type: "todoist-qa", prompt: "Verify and execute: [scribe's formatted commands]. Original intent: [distill's validated plan]")

Step 6: Continue or Finish

After each batch of 5:

• Show running stats: Completed: X | Rescheduled: X | Deleted: X | Someday: X | Kept: X
• Ask: "Continue with next batch? (Y tasks remaining)" via AskUserQuestion

Step 7: Summary

When done (all tasks processed or user stops):

## Triage Complete

| Action | Count |
|--------|-------|
| Completed | X |
| Rescheduled | X |
| Deleted | X |
| Moved to Someday | X |
| Kept as-is | X |

Overdue count: [before] → [after]

Key Principles

• Never assume — always ask the user. Even obviously stale tasks might be important.
• Flag staleness honestly — a task overdue 3+ weeks deserves a direct "is this still real?" prompt. Don't soften it.
• Batch writes — collect decisions for 5 tasks, then send to todoist-scribe in one shot. Don't make API calls one at a time.
• Respect "research" — when the user needs context, get it. The point is informed decisions, not speed.
• Track the Someday project — on first Someday action, check if a "Someday" project exists. Create it if not. Remember the project ID for subsequent moves.