The hardest part isn't the code. It's opening Notion, reading the ticket, figuring out where to start, and not getting pulled into Slack on the way. That fifteen minutes between "I have a ticket" and "I'm writing code" is where ADHD eats the day. Context assembly, remembering what I did yesterday, processing the messages people flagged for me, keeping the board honest — all executive-function-heavy, all low-dopamine, all the kind of work that makes me open a new tab instead.
I've been building Claude Code skills to route around this. Not as productivity hacks or life optimisations — as compensations for a brain that drops balls at the handoff between "I noticed something" and "I did something about it." Six commands that replaced six categories of friction I used to white-knuckle through or just skip entirely.
/triageThe lowest-friction action I can take when I notice a bug is to react with an eyes emoji on a Slack message. That's it. One tap. No context switch, no tab change, no "I'll write that up later."
Later, I run /triage. It scans my Slack for messages I've reacted to with eyes, searches Notion for existing tickets that might already cover the issue, and creates new tickets for anything untracked. The tickets come out scoped and structured — "[iOS] Recipe images not loading on slow connections" — with source links back to the original Slack thread, the relevant code path, and whatever context was in the message.
There's an emoji lifecycle that makes the state visible without me maintaining it. Eyes means I noticed it. Ticket emoji means triage created a ticket. Checkmark means it's processed. I can glance at a Slack channel and know what's been captured without opening Notion.
/workI used to stare at a Notion ticket for five minutes before doing anything. Read the title. Read the description. Open the codebase. Grep for the relevant component. Open Slack to see if anyone said more about it. Check if there was a related PR. By the time I had enough context to start, I'd lost the thread or found something else to distract me.
/work takes a Notion URL and does all of that for me. It fetches the ticket, assesses whether there's enough context, and if not, searches Slack and the codebase to fill the gaps. When it's done researching, it asks me a specific question with options — not open-ended "what do you want to do?" but "the crash is in the image loader — do you want to (a) add a retry or (b) preload on the list screen?"
By the time it asks me something, I'm already in context. The research is done. The decision is multiple-choice, not open-ended. That's the difference.
/groomNotion boards rot. Not because anyone decides to let them rot, but because updating a ticket's status after merging a PR is exactly the kind of low-stakes, low-dopamine task that ADHD deprioritises. The PR gets merged, the code ships, and the ticket sits in "In Progress" for two weeks because no one went back to drag it.
/groom scans recent PRs, cross-references them against Notion tasks, and suggests status transitions. Merged PR linked to a ticket? It suggests moving it to "Ready for QA." Icon doesn't match the status? It flags the mismatch. Ticket marked "In Progress" but no commits in a week? It surfaces that too.
Each suggestion comes individually. I approve or skip one at a time. The skill doesn't batch-update the board behind my back — it presents a stream of small decisions I can process without loading the entire board state into my head.
The board stays honest because the maintenance cost dropped from "open Notion, review every ticket, remember what I shipped" to "run groom, tap yes or no ten times."
/standupStandup is a memory test. What did I do yesterday? ADHD makes that question genuinely hard. Not because I didn't do things, but because episodic recall for mundane work events is unreliable when your working memory already dropped the context.
/standup doesn't ask me to remember. It pulls from Slack messages I sent, GitHub PRs I opened or reviewed, and Notion tasks that changed state. It reconstructs yesterday from artifacts — things I provably did — instead of relying on a recall system that doesn't work.
The output is a short summary I can paste into the standup thread. Accurate, complete, and assembled without me staring at a blinking cursor trying to reconstruct a day I genuinely can't replay in my head.
/data-digest via /loopI run /data-digest on a recurring loop. It pulls per-project metrics from BigQuery — new signups, top events, API costs, error rates, recent feedback — and presents a morning summary across all my projects.
This one isn't about a specific executive function gap. It's about the broader problem of dashboard fatigue. I have analytics for four projects across three dashboards, and the activation energy to open any of them, interpret the data, and context-switch back to whatever I was doing is high enough that I just don't do it most days.
The digest runs unattended. I glance at it over coffee. If something looks off — a spike in errors, a drop in signups, an LLM cost that jumped — I dig in. If everything looks normal, I move on. No dashboards opened, no logins, no context switches.
Looking at these together, three design principles keep showing up.
Reconstruction over recall. Standup and data-digest don't ask me to remember anything. They reconstruct my day from artifacts — messages sent, PRs merged, events logged. Memory is the bottleneck ADHD makes unreliable. These skills route around it.
Context assembly as a service. Work does the research before I start coding. The fifteen minutes of orientation that used to precede every task now happens in the skill, not in my head.
Multiple-choice over open-ended. Work asks "option A or option B?" not "what do you want to do?" Groom presents one suggestion at a time, not the whole board. Small decisions are easier than big ones. Every skill minimises the decision surface at each step.
I didn't design these by reading a paper about executive function. I built them because I kept dropping the same balls in the same places. If you have ADHD and you use Claude Code, start with standup and triage — they compensate for the two most common failure modes: forgetting what you did, and forgetting to write down what you noticed.