Free tool
Sage
The calendar PA I built for my ADHD, because planning is the one thing my brain won't do. Sage reads what people are waiting on you for, works out what actually matters, and books it into the gaps in your calendar as sage-green focus blocks. When a meeting lands on one, it moves it. When you've replied, it clears it.
Built because I can't plan
I have ADHD, and I've stopped being coy about it. Planning is the part it takes from me completely. I can do the work; deciding what to do when, and holding on to everything I've promised people, is what falls apart.
It doesn't help that my calendar is mostly other people's meetings, or that my to-do list never lives in one place. It's an email asking for pricing, a Slack thread where someone tagged me, a Jira ticket, a demo on Thursday that needs prepping, and "I'll send that over after the call", said out loud and written down nowhere. A to-do app doesn't fix any of that, because a to-do app is one more thing to plan, maintain and remember to open.
Sage does the planning for me, and it has made a dramatic difference to how my weeks go. This is what it takes off my plate:
- Deciding what's next. The next thing is already in my calendar, at a time it fits, with a line saying why it's there. No staring at a list working out where to start.
- Keeping a list. There isn't one to maintain. It finds the asks itself, three times a day, and clears them once I've replied.
- Losing track of time. Every task becomes a block with a start and an end, drawn between the meetings it has to fit around. Time blindness has a much harder job when the time is right there on the screen.
- Forgetting what I promised. The "I'll send that over" I said on a call and forgot by the end of it gets picked up from my sent mail and call notes, and once a week it reads a month back for anything that slipped.
- Putting off the small stuff. Two-minute replies get batched into one "Quick hits" block, which is far easier to start than five separate chores.
If your brain works the same way, this was built for you. If it doesn't, it's still a pretty good PA.
Claude reads, a small engine does the maths
The split is deliberate. Claude does the part that needs judgement: reading the email and Slack threads, working out whether someone is actually waiting on you, and tagging which account it belongs to. A plain Python engine does everything that should be exact: scoring, finding the gaps, placing blocks, and comparing the result with what is already in the calendar. Times are computed, never guessed, and a run with nothing new changes nothing.
What comes first
You list your accounts in three tiers. If HubSpot is connected, setup drafts that list from your open deals for you to adjust. Every to-do then gets a score, and the highest score gets time first.
| Factor | Points |
|---|---|
| Task priority | P1 30, P2 20, P3 10 |
| Account tier | Tier 1 +30, Tier 2 +15, Tier 3 or unlisted 0 |
| Deadline | Overdue +25, due by the next working day +20, within three working days +10 |
Anything scoring 60 or more is a must-do: it can run up to an hour past the daily cap, and it shrinks rather than gets dropped. Short tasks share a "Quick hits" block, lunch is protected, and by default it never books more than three hours of your free time in a day. Every block says why it is ranked where it is, so when it gets a call wrong you can see which part to correct.
Living with it
- Done? Grey the block out, put a tick in its title, or press the "Mark done" link inside it. There is one link per item, even inside a quick-hits block.
- Not now? Delete the block and that to-do is snoozed for three working days.
- Wrong time? Move it, and Sage leaves it exactly where you put it.
- Replied already? It notices on the next run and clears the block.
The Mark done link normally opens a pre-filled email to yourself. If you'd rather it were a single tap, the download includes a tiny Google Apps Script web app that you deploy under your own account: press the link, a tab says "Marked done", and that's it.
It never speaks for you
Sage writes only to your own calendar, and only to events it created. It never sends an email or a Slack message, never adds attendees, and never accepts or declines anything. Everything it reads is treated as data rather than instructions, so an email cannot tell it to do things.
Colleagues can see the blocks
That's on purpose: a calendar full of "Focus" blocks tells people you're heads-down rather than free. Blocks follow your calendar's sharing, so Sage writes their descriptions for that audience, with no deal values and no opinions about people. Anything about hiring, people matters or something you were told in confidence is marked private, as is the daily banner. If you'd rather nobody saw any of it, choose private at setup and colleagues only ever see "busy".
Not on Google?
Sage is tested every day on Google Calendar, Gmail, Slack, Jira and HubSpot. Outlook, Teams, Linear, Asana, Salesforce and friends are supported by mapping rather than by new code: the engine only ever says what should change (a title, a time, a colour, private or not, busy or free), and Claude carries that out through your own connectors. Setup works out how your tools say the same things, and for a calendar other than Google it tests the round trip with one throwaway event before it touches anything else.
To be straight about it: those paths are untested until someone runs them. If you're that someone, I'd love to hear how it goes.
Installing it
-
Download the zip from the release page, unzip it, and move the
sage-calendar-pafolder into~/.claude/skills/. Or clone the repo straight there. - Connect your calendar and email in Claude (Google Calendar and Gmail are the tested pair). Chat is strongly recommended; a task tracker, a CRM and call notes are optional.
- In Claude Code, say "set up Sage". It works out who you are, asks three questions, drafts your account tiers (from your CRM, if one's connected), and does a dry run that writes nothing until you say yes.
It will then offer to schedule itself and to add the permissions unattended runs need.
Both are optional and both ask first. Your settings and history live in
~/Documents/Sage, outside the skill, so updating it never touches them.
The honest caveats
- This is a personal project and it is experimental. I use it every weekday and it does the job, which is a different claim from it being finished.
-
It reads your email and chat. Only through your own Claude
connectors, and it sends nothing anywhere else, but you should know that is what it does
before you switch it on. Its run logs contain task context, so treat
~/Documents/Sagelike your inbox. - Its priorities are Claude's judgement. The scores are exact, but whether an email is really an ask is a reading of the text. The reasoning line on every block is there so you can catch it out.
- Scheduled runs need the Claude desktop app on a Mac, open at the time. On demand, it works in any Claude Code session. It won't work as a skill uploaded to claude.ai on the web, because it needs local Python and a folder that persists between runs.
- It assumes one timezone, weekday working hours, and English in your email and Slack.
If you try it and it gets something wrong in a way I haven't thought of, I'd like to know. Issues on the GitHub repo are the easiest way, or there is a contact link on the front page.
