Free kit · from the video

The Hands-Off Claude Kit

Every copy-paste prompt from the goals, loops and routines video — the exact ones running my business — plus the cheat sheet that tells you which one to use.

No spam. Unsubscribe anytime.

The Hands-Off Claude Kit

Stop babysitting AI.
Hand the job over.

Goals hand off the checking. Loops hand off the timing. Routines hand off the repeating. Every prompt below is copy-paste ready.

start here ↓

Which one do I need?

Ask yourselfUseBecause
Does it have a finish line?GOALan independent checker holds the line
Am I waiting on something today?LOOPwatches on a timer, dies with the session
Do I repeat this every day or week?ROUTINEsaved prompt, fresh run, on a schedule
Have I taught Claude how already?+ SKILLdrop the skill into any of the three

Goals — set the finish line

You write what "done" looks like. Claude works, a separate checker judges each turn against YOUR definition, and it goes again without you typing anything. Big goals use all six parts:

TASK

the actual work — build it, rewrite it, categorise it. Not just "check it".

WHY

the stakes, so Claude makes trade-offs your way

OUTCOME

the thing that should exist at the end — a file, a page, a finished list

CONSTRAINTS

the rules + the rubric to score against ("that file is the definition, not your taste")

VERIFICATION

what to print each turn so the checker has evidence — tables, counts, scores

STOP

a turn cap so it can't spin forever

GOAL · TIDY A MESSY EXPORT
/goal every transaction in my Xero export is categorised against
my chart of accounts. Print the count of uncategorised rows each
turn. Done when it's zero. Stop after 6 turns.
GOAL · NO PLACEHOLDERS LEFT
/goal every section of onboarding-template.md is filled with this
client's actual details from the discovery notes. No placeholder
text, no "TBD". List the remaining gaps each turn. Stop after
5 turns.
GOAL · FIX MY EMAIL SUBJECT LINES (FULL STRUCTURE)
/goal
TASK: Rewrite the draft subject lines in subject-lines.md until
every one passes my rubric below. Edit the file directly — keep
what already passes, rewrite the failures.
WHY: This goes to my warmest list. Weak subject lines waste it.
OUTCOME: The file holds finished lines, all passing every check.
CONSTRAINTS: The rubric is the definition of good — not your
taste. Rubric: under 45 characters, all lowercase, no emoji, no
hype words (game-changing, unlock, don't miss out), curiosity 9+
(could they guess the content without opening? then it's under 6),
conversion 9+ (the click must be the only way to resolve it).
VERIFICATION: Print the full scoring table each turn — line,
chars, lowercase, banned words, curiosity /10, conversion /10,
verdict. Justify every 9+ in one sentence. Mark a line STUCK
after 3 failed attempts.
STOP: After 6 turns, or when every line reads PASS or STUCK.
GOAL · BUILD A LANDING PAGE (FULL STRUCTURE)
/goal
TASK: Build a one-page landing page for my free guide at
landing/index.html — hero, pain, what's inside, proof, FAQ,
email signup.
WHY: This page turns social traffic into subscribers. Every weak
section costs signups.
OUTCOME: A finished page that renders cleanly and reads like me.
CONSTRAINTS: ONE call to action, repeated three times, nothing
else to click. Headline names the OUTCOME the reader gets, not
the product. Mobile-first.
VERIFICATION: Each turn print the checklist — section, present
Y/N, copy score /10 naming the criterion that earned it, CTA
count. Then open the page and confirm no horizontal scroll.
Done when every section is present at 9+.
STOP: After 6 turns.
The pattern: if you can count it, you can goal it. And the checker only judges what's on screen — that's why every prompt says "print the table each turn".

Loops — the sitting-and-refreshing tool

Re-runs your prompt on a timer while your session is open. Waiting on a reply, watching launch-day sales, checking if the send finished — that's loop territory.

LOOP · LAUNCH-DAY WATCHER
/loop 30m check my ticket sales sheet. If any tier crossed 80%
sold since last check, flag me and draft an urgency post in my
voice. Otherwise just say "no change".
LOOP · WAITING ON A HUMAN
/loop 20m check my inbox for anything from the venue about the
AV setup. If it's arrived, summarise it and draft a reply.
Otherwise just say "nothing yet".
LOOP · NO TIMER — CLAUDE PACES ITSELF
/loop check whether my email sequence has finished sending. When
it has, pull the open rate and click rate and tell me.
House rules: loops only run while the session is open · recurring loops expire after 7 days (a feature — forgotten loops can't run forever) · press Esc to stop · always include the boring answer ("otherwise just say no change") so a non-event costs two words, not two paragraphs.

Routines — run it without you

A saved prompt on a schedule. Fresh session every run, whether you're at the desk or at the beach. Set these up as scheduled tasks in the Claude desktop app (local tasks can see your files and connectors).

ROUTINE · BRAIN DUMP PROCESSOR — WEEKDAYS 7AM
Read my Brain Dump page. Sort each item into the right project,
turn anything actionable into a task with a due date, and clear
the dump. If something's too vague to sort, leave it and tell me.
If the page is empty, say so and stop.
ROUTINE · MONDAY MONEY CHECK — 8AM
Pull last week's calendar and my invoice list. Flag any client
work that happened but hasn't been invoiced. Post it to Slack as
a short list. If nothing's outstanding, say so — do not invent
numbers.
ROUTINE · SEO HEALTH CHECK — MONDAYS 7AM
Go through every blog post on my website. For each one check:
title under 60 characters, meta description present (120-155
chars), exactly one H1, the target keyword in the first 100
words, at least 2 internal links, alt text on every image.
Print the table — page, each check, PASS/FAIL. Draft fixes for
the 3 worst pages as a report for me — never publish changes
yourself. If a page won't load, mark it UNREADABLE and move on.
A failure instruction

"If there's no data, say so and stop — do not invent numbers."

A don't line

"Never send. Draft only. Nothing publishes without me."

A boring answer

"If nothing happened, just say so." Two words, not two paragraphs.

Every unattended prompt needs those three lines. They're the difference between a routine you trust and one you have to check up on (which defeats the point).
where it gets powerful ↓

Stack them

COMBO 1 · A SKILL AS THE LOOP
/loop 30m /competitor-watch
COMBO 2 · A ROUTINE THAT USES A SKILL — WEEKDAYS 6AM
Check my inbox for anything tagged "needs-reply" from the last
24 hours. Use my email-drafter skill to draft replies in my
voice. Never send — drafts only. If the email is about pricing
or a complaint, don't draft at all, just flag it for me.
COMBO 3 · A ROUTINE WITH A FINISH LINE — WEEKDAYS 7AM
Process everything in my Brain Dump page. Keep working until the
dump page is empty and every item has landed somewhere — print
the before and after counts so I can see it. If you can't place
an item after 3 tries, park it in "unsorted" and tell me why.
Then stop.
COMBO 4 · ROUTINE FINDS → GOAL FIXES
/goal
TASK: Fix every failed check on the 3 worst pages from Monday's
SEO report. Edit the files directly.
OUTCOME: All 3 pages pass all 6 checks.
CONSTRAINTS: Don't change what a post says — only titles,
excerpts, links and structure. Nothing gets published without me.
VERIFICATION: Re-run the same 6 checks each turn and print the
before/after table. Done when the 3 pages show zero fails.
STOP: After 5 turns.
The one-line version: skills know HOW · goals define DONE · loops handle WHEN (today) · routines handle WHEN (forever). You can't type /goal inside a routine — write the finish line into the prompt in plain words instead.

Become the AI expert in your industry

The Wright Mode membership is where women founders stop consuming AI content and become the person everyone in their industry asks about it. Fortnightly co-work and Q&A, monthly live AI-system builds, and every template and prompt library I make — including the systems in this kit, built live.

Join Wright Mode →