LLM Gtd — AI skill for Claude Code

LLM-GTD

**A portable GTD skill for LLM agents.**

Not a todo app. Not another productivity prompt. LLM-GTD is a trusted external system that lets Claude Code, Cursor, and Codex help you capture, clarify, update, organize, choose, and review your work without turning your life into a pile of half-processed tasks.

The installable Codex plugin is named `llm-gtd`. The legacy skill package remains `gtd-harness` for Cursor and manual installs.

See it in one example

You don't hand it a task — you hand it an **intention**:

**You:** I want to help a family member try a class **LLM-GTD:** Makes it a **project**; first action → *"talk for 10 min — do they want to try it, what constraints matter, how often could this happen…"*. It does **not** enroll, schedule, or contact anyone.

A while later you report back:

**You:** Talked it through — they are interested, two sessions/week, ~1h each **LLM-GTD:** Checks the talk off; **advances the project** to *"shortlist 2-3 options — slots, price, distance, trial terms"*; logs the facts as support material; flags your "~1h" as **needs-confirmation**.

A project with nothing in play (no next action, no waiting-for, no dated tickle) is a stalled promise. LLM-GTD keeps the promise **live across days** — you supply judgment and report reality; it does the bookkeeping. → [full walkthrough](docs/demo.md)

The Short Version

Most AI productivity workflows fail for the same reason most human GTD systems fail: the inbox fills, the next action is vague, the project has no current move, and the weekly review gets skipped.

LLM-GTD gives the agent a skill-backed trusted system:

• plain Markdown state in memory/gtd/
• a full GTD workflow, not just inbox triage
• one /gtd router plus seven workflow commands: init, capture, clarify, update, organize, engage, review
• one shared trusted system across Claude Code, Cursor, and Codex
• automatic calendar-provider writes for complete schedule items, with fail-clo