Gtdskill

GTD Processing Engine

Manage Ryan's commitments by bridging two spaces: Context (thinking) and Reminders (action).

Load [references/gtd-rules.md](references/gtd-rules.md) before processing any items.

The Two Spaces

Dev-logs [ ]/[x]  =  AI's workspace     → tasks completed in dialogue, AI executes
Reminders flags   =  Ryan's action list  → tasks completed in physical world, only Ryan can do

These spaces are independent. Never mix them.

Reminders is the confluence of two directions:

AI extracts human actions from Context ──→  Reminders  ←── Ryan captures ideas to 收集箱
            (top-down)                                        (bottom-up)

Mapping: Context ↔ Reminders

• ~/context/01Projects/[name]/ ↔ Reminders list with same name
• ~/context/02Areas/[name]/ ↔ Reminders list with same name
• Reminders "收集箱" = GTD Inbox (unprocessed captures)
• 59 Reminders lists already mirror Context structure

Sensing Both Spaces

# Reminders side
~/.claude/skills/reminders/scripts/reminders.sh list "收集箱"
~/.claude/skills/reminders/scripts/reminders.sh list ""
~/.claude/skills/reminders/scripts/reminders.sh list-lists

# Context side — use Read/Glob/Grep
# ~/context/01Projects/CLAUDE.md — project index
# ~/context/02Areas/CLAUDE.md — area index
# Project details: ~/context/01Projects/[name]/dev-logs/DEVELOPMENT_LOG.md
# Area details: ~/context/02Areas/[name]/CLAUDE.md

Load what's relevant to the conversation, not everything.

Processing an Item

For any item (from inbox or identified from Context):

1. What is it? — Actionable / reference material / idea seed / trash
2. Where does it belong? — Match to existing project/area, or new
3. Who does it? — Ryan (physical world) / AI (dialogue) / together
4. Atomic? — Can it be done in one step? If not → decompose

When matching to a project/area, read its current state first (dev-logs or CLAUDE.md) to check if redundant, already in backlog, or genuinely new.

Routing

| Result | Destination | Command | |--------|------------|---------| | Ryan + atomic | Reminders list, flagged + due date | reminders.sh edit "" --move-to-list "" --flagged --due | | Ryan + needs decomposition | Decompose first, then route each sub-task | — | | AI does it | Dev-logs task + Reminders "启动AI做X" | Write to dev-logs + reminders.sh edit | | Together | Reminders with note + dev-logs if project-related | reminders.sh edit "" --move-to-list "" --notes "和AI一起做" --flagged --due | | Reference material | Context (03Resources or Area) | Create/update file + reminders.sh delete "" | | Redundant | Delete | reminders.sh delete "" |

Reminders Commands

reminders.sh list ""          # list items
reminders.sh add "" --list ""  # add item
reminders.sh edit "" --move-to-list "" --flagged --due  --time  --notes ""
reminders.sh delete ""            # delete item
reminders.sh search ""            # search across lists
reminders.sh list-lists                  # list all lists

Flag = atomic action = next physical action. Always set a due date when flagging.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: RyanWangFun
• Source: RyanWangFun/claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.