gtd-process

name	gtd-process
description	Use when processing, triaging, or clarifying GTD inbox items into projects and context-list actions following the GTD clarify and organize workflow.

Process GTD inbox items into projects or actions. The agent infers categorization, project assignment, priority, time estimate, and due date for each item — then presents a single confirmation for the user to approve or modify.

CLI Tool

Run the Swift source directly (no build step required). <plugin-root> is the installed plugin directory — the directory that contains scripts/ and codex-skills/ (two levels above this skill's folder; in a repository checkout it is productivity-skills/). Resolve it to an absolute path and substitute it in every command:

swift <plugin-root>/scripts/productivity-cli.swift <command>

Step 1: Determine Processing Mode

Default: process all items sequentially until the inbox is empty, every item has been presented this session, or the user selects "Stop" on the per-item confirmation.

Only switch to single-item mode if the user explicitly asks for it (e.g. "just one item", "single item", "process the first one" — non-exhaustive). In that case, process the first item and exit via Step 6.

Step 2: Read Inbox and Gather Context

1. Read ~/.claude/productivity-skills/inbox.md (shared with the Claude Code version of this skill — keep this exact path)
2. If file or directory does not exist, create directory and file:

   mkdir -p ~/.claude/productivity-skills
   

   Then write an empty inbox file with header # Inbox and a blank line.
3. If empty, inform user: "Inbox is empty. Nothing to process." and exit.
4. Display numbered list of items.
5. Ensure required lists exist, creating any that are missing:

   swift <plugin-root>/scripts/productivity-cli.swift reminders lists
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "Projects"
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "@quick"
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "@1pomo"
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "@2pomo"
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "@deep"
   swift <plugin-root>/scripts/productivity-cli.swift reminders create-list "@agent"
   

6. Fetch existing projects for context:

   swift <plugin-root>/scripts/productivity-cli.swift reminders incomplete "Projects"
   

Step 3: Analyze and Propose

For each inbox item, analyze the text and infer:

1. Category: Is this a new project (multi-step outcome), a project action (belongs to an existing project), or a single action (standalone task)?
   • If existing projects were fetched, check if the item relates to any of them
2. Project assignment: If it's a project action, which project does it belong to?
3. Action title: Clean up the inbox text into a clear action title
4. Task type: Is this human-centric or agent-centric?
   • Agent-centric indicators: "generate", "draft", "analyze", "research", "summarize", "review code", "run tests", "scan", "convert", "process" — tasks that produce an artifact the user monitors and checks later
   • Human-centric indicators: phone calls, meetings, physical tasks, decisions requiring real-time judgment
   • Default to human-centric when ambiguous
5. Time estimate (human-centric only): Infer from complexity — @quick (< 25 min), @1pomo (25 min), @2pomo (50 min), @deep (90+ min). Skip for agent-centric tasks.
6. Priority: Infer from urgency cues in the text (default: Medium)
7. Due date: Infer from any time references in the text (default: one week from today)

For new projects, also infer:

• Project name: {CamelCaseSummary}-{YYYYMMDD}
• End goal: What "done" looks like based on the item text

Present the proposal and ask the user a single question: "Confirm or modify this processing:"

Human-centric example:

Item: "Call dentist to schedule appointment"

Proposed:
  Type: Single Action
  Task type: Human
  Action: "Call dentist to schedule appointment"
  List: @quick (~15 min)
  Priority: Medium
  Due: 2026-05-04

Confirm or modify?

Agent-centric example:

Item: "Analyze Q1 sales data and generate summary report"

Proposed:
  Type: Single Action
  Task type: Agent
  Action: "Analyze Q1 sales data and generate summary report"
  List: @agent
  Priority: Medium
  Due: 2026-05-04

Confirm or modify?

New project example:

Item: "Research vacation flights by end of month"

Proposed:
  Type: New Project
  Project name: VacationResearch-20260324
  Goal: Flights researched and booked
  Priority: Medium
  Due: 2026-03-31
  First action: "Search flight comparison sites" (@1pomo, Human)

Confirm or modify?

Options: "Confirm", "Modify", "Skip", "Stop"

• Confirm: Execute the proposed processing (create project/action via CLI), then continue
• Modify: User provides corrections, then execute with modifications, then continue
• Skip: Leave item in inbox, advance to the next unpresented item
• Stop: Leave current item in inbox and exit the processing loop entirely — unlike Skip, no further items are presented

For Skip and Stop, do not perform Step 4 or Step 5 — proceed directly to Step 6.

Step 4: Execute Processing

Based on the confirmed or modified proposal:

For new projects:

1. Create project in Projects list:

   swift <plugin-root>/scripts/productivity-cli.swift reminders create \
     --title "ProjectName-YYYYMMDD" \
     --list "Projects" \
     --priority 5 \
     --notes "Goal: [end goal]" \
     --due "YYYY-MM-DD 17:00"
   

2. Create first action in the appropriate context list (use the inferred list from Step 3):

   swift <plugin-root>/scripts/productivity-cli.swift reminders create \
     --title "Action title" \
     --list "<inferred list>" \
     --notes "#{ProjectName-YYYYMMDD}" \
     --priority 5
   

   Use @agent if the first action is agent-centric, otherwise use the inferred time-based list (@quick, @1pomo, @2pomo, @deep).

For project actions (use the inferred list from Step 3):

swift <plugin-root>/scripts/productivity-cli.swift reminders create \
  --title "Action title" \
  --list "<inferred list>" \
  --notes "#{ProjectName-YYYYMMDD}" \
  --priority 5 \
  --due "YYYY-MM-DD 17:00"

For single actions (human-centric — use the inferred time-based list):

swift <plugin-root>/scripts/productivity-cli.swift reminders create \
  --title "Action title" \
  --list "<inferred list>" \
  --priority 5 \
  --due "YYYY-MM-DD 17:00"

For single actions (agent-centric):

swift <plugin-root>/scripts/productivity-cli.swift reminders create \
  --title "Action title" \
  --list "@agent" \
  --priority 5 \
  --due "YYYY-MM-DD 17:00"

Omit --due only if user explicitly requests no due date. Omit --priority if None (0).

Step 5: Remove from Inbox

Edit inbox.md to remove the processed item.

Step 6: Continue or Exit

Track which inbox items have been presented in this session (by their original text). Exit (show summary of processed items) when any of these are true:

• User selected "Stop" in Step 3
• Inbox is empty
• Every item currently in the inbox has already been presented this session (each was either processed or skipped) — this prevents an infinite loop when the user skips every item
• Single-item mode was requested in Step 1 and the first item is done

Otherwise, return to Step 2 to refresh the inbox contents and the existing-projects context (Step 4 may have created a new project), then propose the next unpresented item in Step 3. Do not ask the user whether to continue.

Reference

Context lists (human): @quick (< 25 min), @1pomo (25 min), @2pomo (50 min), @deep (90+ min)

Context list (agent): @agent (no duration — async, monitor progress)

Priority values: 1=High, 5=Medium, 9=Low, 0=None

Date format: yyyy-MM-dd HH:mm (default time 17:00)

Project naming: {CamelCaseSummary}-{YYYYMMDD}

Project linking: #{ProjectName-YYYYMMDD} in action notes field

Guidelines

• Every project must have a clear end goal
• Create missing reminder lists automatically before creating reminders
• Project actions inherit priority from the project unless overridden
• Infer as much as possible from the inbox item text — minimize user interaction
• Present one confirmation per item, not multiple sequential questions