Getting Things ADHDone

An Obsidian + Claude Code setup for running a second brain when your working
memory won't cooperate. Less "productivity system," more a body double -
a working partner that sits with you, holds the list you can't hold in your
head, and nudges you toward starting rather than organising forever.

Why this exists

Most PKM (personal knowledge management) advice assumes you'll maintain the
system. If you have ADHD, the system is usually the first thing to fall over -
not because the ideas are bad, but because "spend 20 minutes weekly reviewing
your task database" is itself an executive-function tax you don't have spare
capacity for.

This repo is the alternative: Claude Code does the maintenance (finding
carried-over tasks, catching untagged loose ends, tidying at day's end) so the
human only has to do two things - write things down roughly, and answer a
short prioritisation question when asked. The structure stays flat and
low-ceremony on purpose; the "skills" below do the retrieval work instead of
asking you to keep it organised yourself.

What's in here

• CLAUDE.md - project instructions for Claude Code: vault structure,
  tagging conventions, task/date conventions. Drop this at the root of your
  own vault and edit to taste.
• vault-skeleton/ - an example folder structure (Notes, Projects,
  Archive, Templates, Daily notes, etc.), for starting a vault from scratch.
  If you already have a vault, skip it and map the conventions onto your
  existing folders instead - see that folder's own README for both paths.
• templates/ - Templater templates for daily notes and generic/company
  notes.
• skills/ - Claude Code skills that do the body-doubling work. Every
  skill here is either standalone (does one job, callable on its own) or
  an orchestrator (does no work itself, just calls other skills in
  sequence) - never both, and never a skill that does several unrelated jobs
  in one go. New skills added later should keep to this split rather than
  growing into a do-everything script. Currently:
  • start-the-day - orchestrator only: names the chat for the day (Claude
    desktop app only), optionally checks in about your task inbox and any
    other capture inboxes, opens today's Daily note, asks whether you want a
    stale-task review (default no) and runs surface-stale-tasks only on a
    yes, then runs pick-meetings-to-prep. Doesn't do any of that work
    itself.
  • surface-stale-tasks - flags to-dos that may have gone stale (7+ days in
    your Today list, 21+ days anywhere else), summarised rather than dumped
    as a list. Standalone - run it any time you want a staleness check
    without the rest of the morning ritual.
  • pick-meetings-to-prep - asks which of today's meetings you want to
    prep for, then hands each one to prep-meeting. No calendar access
    needed by default; the skill's own notes cover an optional automated
    calendar read on macOS, and why the author dropped it.
  • prep-meeting - given a meeting name, finds or creates a dated meeting
    note and pulls in matching tasks by tag. Standalone - run it for any
    meeting, any time of day.
  • update-tasks - refreshes a human-readable snapshot note of your task
    manager's open to-dos, if you keep one. Standalone.
  • link-task - on request only, creates a two-way link between a specific
    task and a specific note: a deep link to the task added to the note, and
    a link back to the note appended to the task. One of two deliberate,
    narrow exceptions to the "read-only against your task manager" rule the
    other skills here follow (the other is end-the-day adding an inbox item
    on a per-item yes) - see the skill's own guardrails.
  • All of the above work whether you track tasks in the vault itself or in a
    separate task manager - see start-the-day's notes on that choice, plus
    an example Things3 integration under skills/things3-export/.
  • end-the-day - once a day, scans what you touched today for loose ends
    • action items that aren't actually tracked anywhere, unresolved
      questions, duplicates - and walks through fixes one at a time, offering
      to add untracked actions to your task inbox on a per-item yes. Never
      silently edits anything. Also runs log-completed-tasks, catches real
      work on "sparse" days that never touched the task manager, runs
      suggest-brag-items over the same files, and finishes with a
      verification pass that re-reads everything it claims to have changed.
  • end-task - the lighter per-chat counterpart: before you close a task
    chat, checks that Claude's own memory files reflect what the chat just
    did, so the next chat starts from accurate context. Standalone.
  • log-completed-tasks - writes what your task manager shows as completed
    today (or on a given date) into the Daily note, as a plain factual list.
    Standalone, or run from end-the-day.
  • suggest-brag-items - looks back over what you wrote today for things
    worth remembering you did, and drafts a short, linked bullet for each
    one to add to a running "brag list." Added because it's easy to end a
    day only remembering what's still unfinished - this exists to counter
    that, not to replace your own judgment about what counts as worth
    keeping.
  • search-mail - on request only, searches and reads back email via
    AppleScript against Apple Mail rather than an OAuth mail connector -
    local app access instead of a connector, useful if a connector isn't
    available or vetted for the account in question. Read-only, standalone.
• docs/vault-setup.md - the mechanics doc: how templates, folders, and
  the 1:1 tagging system are wired together, plus known failure modes
  (macOS smart quotes breaking Templater scripts, stale frontmatter blocks,
  tag substring collisions).
• docs/obsidian-plugin-setup.md - which Obsidian community plugins this
  depends on and the settings that actually matter (Templater's template
  folder mapping, for one - the Tasks plugin and its global filter are only
  relevant if you're tracking tasks inside the vault rather than a separate
  task manager, see below).

Requirements

• Obsidian with the plugins listed in
  docs/obsidian-plugin-setup.md - Templater and Dataview are required. The
  Tasks plugin is only needed if you're tracking tasks inside the vault
  rather than in a separate task manager (Things3, Todoist, Reminders.app,
  etc.) - see start-the-day's notes on that choice.
• Claude Code pointed at your vault
  directory.

Terminal or desktop app?

Claude Code runs in a terminal or in the Claude desktop app's Code tab. The
author started in the terminal and moved to the desktop app, which suits
this workflow better:

• Formatted replies are easier to read than terminal text
• The sidebar makes it easy to keep one "daily" chat open all day and fork
  bigger tasks into their own chats
• Claude can rename chats and file them into sidebar groups, which
  start-the-day and end-the-day use to name each day's chat and file it
  away
• Claude can read what you ran in the app's Terminal panel

Two caveats:

• Chats started in the terminal don't appear in the app's sidebar
  automatically. To bring one across, run /desktop in the terminal, which
  opens the chat in the desktop app and closes the terminal session, or type
  /resume in the app to pick from your terminal chats. (/desktop needs
  macOS or x64 Windows and a Claude subscription sign-in, not an API key.)
• Commands that open interactive terminal dialogs (such as /permissions or
  /config) aren't available in the app, so run those from a terminal
  session

Setting it up

Quicker route: if you already have Obsidian and Claude Code installed,
navigate to your vault root and run:

git clone https://github.com/pholbo/getting-things-adhdone /tmp/gta
/tmp/gta/setup.sh .
claude

then type /setup inside Claude Code. It'll ask a few questions (new vault
or existing, how you track tasks) and copy in CLAUDE.md, templates, folder
structure, and the right skill set for your answers. You'll still want to
install the Obsidian plugins below and skim docs/vault-setup.md afterwards,
but this gets you to a working starting point first.

Manual route:

1. Install and configure the Obsidian plugins per
   docs/obsidian-plugin-setup.md first - several of the pieces below
   silently do nothing if a plugin setting is off.
2. Starting from scratch? Copy vault-skeleton/ into your new vault. Already
   have a vault? Skip it and add whichever pieces you're missing (a Daily
   notes folder, a Projects/Archive split) to your existing structure -
   see vault-skeleton/README.md for both paths.
3. Copy CLAUDE.md to your vault root and adjust the folder names/emoji in
   its structure table to match whatever you ended up with (or drop the
   emoji entirely - it's cosmetic).
4. Copy templates/ into your vault's Templater templates folder, and
   skills/ into .claude/skills/ inside your vault.
5. Read docs/vault-setup.md for how the templates, tags, and dashboards
   are meant to fit together, then adapt rather than adopt wholesale - this
   reflects one person's workflow, not a universal standard.
6. Start with /start-the-day each morning and see what sticks.

Philosophy

• Manual over automated, by default. Cron jobs and push notifications
  are easy to build and easy to start ignoring. On-demand skills you
  trigger yourself stay legible - you always know why something happened.
• Never silently edit. end-the-day reports what it'd change and asks
  before touching anything, one decision at a time rather than a wall of
  text.
• Flat over nested. Tags, folders, and structure stay as shallow as
  they can, because a system with a learning curve is a system that stops
  getting maintained.
• One skill, one job. A skill either does a single, standalone thing, or
  it's a thin orchestrator that calls other skills in sequence - never a mix
  of both. start-the-day used to be one skill doing several jobs; it's now
  an orchestrator over surface-stale-tasks, prep-meeting, and
  update-tasks, each of which is independently useful on its own. Splitting
  this way keeps each skill's job statable in one sentence, and lets you run
  a piece of a ritual without the whole thing.

Related

• i-have-adhd - a Claude Code skill
  that keeps the model's own output ADHD-friendly (answer first, no burying
  it under preamble). Complements this repo well: this is about the vault
  Claude works in, that's about how Claude talks back to you. Used alongside
  the skills here.

License

MIT - see LICENSE.