DEV Community: RoutineKit

Before I let an agent touch the repo, I write 4 lines. What's in yours?

RoutineKit — Wed, 09 Sep 2026 00:49:04 +0000

Every time I hand work to an AI agent — a coding agent, a chat window, whatever — the output quality tracks one thing almost perfectly: how much context I bothered to write down before I started typing the actual request.

So I stopped improvising. Now there are four lines at the top of whatever file or thread I'm working in, and I write them before the first prompt:

WHO:    who this is for, and what they already know
DONE:   what "finished" looks like, in one sentence
NEVER:  the two or three things that must not happen
NEXT:   the single next action after this task ships

That's it. Four lines, maybe thirty seconds.

WHO kills the biggest failure mode — the agent writing for a generic reader instead of the specific one. "Senior backend dev, knows our auth stack, has never seen this service" produces different code and different comments than nothing at all.

DONE is the one people skip, and it's the one that saves the most time. If I can't write the finish line in a sentence, the task is actually two tasks and the agent is about to guess which one I meant.

NEVER is where the scar tissue goes. Mine usually reads something like "don't touch the migration files, don't add a dependency, don't refactor anything I didn't ask about." Every line in there exists because it happened once.

NEXT is not for the agent at all. It's for me, forty minutes later, when I come back with the context evaporated. Writing the next action down while I still have the whole problem loaded is the cheapest thing I do all day.

The interesting part is what happened to the prompts themselves. They got shorter. Once the constraints live above the task, the request is just the request — no more paragraph of throat-clearing about what I really meant. And when the output is wrong, the diff between what I got and what those four lines say is usually the whole diagnosis.

I don't think the specific four are magic. WHO/DONE/NEVER/NEXT is just what survived a few months of trimming. The general shape — a tiny fixed header you write before delegating, in the same place every time — seems to be the thing that matters.

Which is why I'm curious about yours.

Do you write anything down before you delegate to an agent, or do you go straight into the prompt?
If you have a header like this, what's in it that isn't in mine?
And for the people who tried structured prompting and dropped it — what made it not worth the thirty seconds?

Genuinely asking. I've got a sample size of one here.

I put a free Module B prompt pack on Product Hunt today

RoutineKit — Tue, 01 Sep 2026 08:27:54 +0000

Blank-page proposals are usually a prompt shape problem, not a model problem.

The free Module B I shipped today is one content-pipeline slot:

role
constraints
output format
a filled example
a stop checklist so the slot ends

If you sell services and still lose half a day to one draft, grab the free sample and tell me which part still balloons.

PH: https://www.producthunt.com/products/routinekit-free-sample
Free sample: https://routinekit.gumroad.com/l/alxurc

The proposal went quiet. Here is a 3-touch cadence that stays human.

RoutineKit — Wed, 26 Aug 2026 04:21:47 +0000

The proposal went quiet. Here is a 3-touch cadence that stays human.

You did the hard part.

You had the call. You wrote the proposal. You hit send. They said I’ll get back to you — or they said nothing, which is the same thing with worse manners.

Then the thread died.

Most solos vanish “to give space,” or they just check in until it feels like spam. What you need is a cadence: three touches with different jobs, then a clean exit — only for people who already engaged.

Why “just checking in” fails

A check-in asks the other person to do work — remember the offer, reopen the file, invent a status. Busy buyers don’t owe you that.

A useful follow-up does one of three jobs:

Make reply easy (yes / not now / no)
Add a small, relevant asset (not a new pitch deck)
Close the loop so you both can stop thinking about it

If a message does none of those, don’t send it.

The 3-touch cadence (day 3 / 7 / 14)

Count from the last their message or from the day you sent the proposal — pick one rule and stick to it. I count from proposal-sent.

Touch 1 — Day 3: the soft bump

Job: lower the effort of a reply.

Assume they meant to answer and got buried. Keep it short. Remind them what you sent in one line. Offer two or three labeled options.

Example shape:

Quick bump on the [project] proposal from Tuesday — 6 pages, [scope in 8 words].

If useful, reply with 1 / 2 / 3:

1) Let’s talk next step

2) Not this month

3) Not a fit

No essay required.

No guilt. No “I know you’re busy.” They know.

Touch 2 — Day 7: the value-add

Job: give them a reason to reopen the file that isn’t “I want a yes.”

Add one useful thing tied to their problem: a one-page risk list, a revised timeline option, a single example of the deliverable, a scoped-down version if budget was the hesitation. One attachment or one paragraph. Not a new proposal.

If you have nothing real to add, skip it or wait for touch 3. Fake “value” is just pressure with a file attached. AI can draft this if you feed the original proposal, the real objection, and a length cap — then delete anything you wouldn’t say on a call.

Touch 3 — Day 14: close the loop

Job: end the open loop for you.

This is the message most people skip because it feels final. That’s why it works. You’re not punishing them. You’re taking the decision off both calendars.

Example shape:

I’ll close the loop on the [project] proposal so it doesn’t sit in your inbox.

If you want to pick it up later, reply “reopen” and I’ll treat it as a new week.

Otherwise I’ll assume it’s a no and wish you well on [their stated goal].

Then stop. Move the row to Closed. Do not send a fourth “really just checking.”

Closing is a professional courtesy. It also keeps your pipeline honest. A maybe that is 19 days old is a no you’re paying rent on.

Track status, not feelings

A cadence dies if it lives in your head. Four statuses are enough:

Quiet — proposal out, no useful reply
Waiting — they named a date (“after the 15th”)
Nudged — you sent a touch; clock running to the next one
Closed — they said no, or you closed the loop

Next follow-up date is a field, not a vibe. Monday you open the due rows. You do not reread the whole graveyard.

Notion, a spreadsheet, or a paper list all work. The tool matters less than the rule: no touch without a next date, no fourth touch.

Tone rules (so you don’t sound needy)

One ask per message
No fake urgency (“another client is waiting on this slot”) unless it’s true
No essays about how much time you spent
Match their channel: if they lived in email, don’t suddenly DM three times
Read the draft out loud. If you wouldn’t say it to a peer, rewrite

Warm is specific and brief. Needy is long and circular.

Where AI belongs

Use a model to turn your notes into a first draft: lead name, what you sent, what they last said, which touch this is (3 / 7 / 14), and “max 90 words.”

Do not ask it to “write a high-converting follow-up.” It will add scarcity and compliments you don’t mean. You are not running a cold sequence. You are finishing a conversation you already started.

Soft CTA

If you want a free taste of the structured-prompt + checklist style I use for weekly work:

https://routinekit.gumroad.com/l/alxurc

If you want the follow-up pack this cadence lives in — Notion schema, day 3 / 7 / 14 templates (email + DM), and 3 AI prompts for quiet leads (not cold strangers) — that’s the Quiet Lead Follow-Up System ($9):

https://routinekit.gumroad.com/l/chdppw

No reply-rate promises. Practical process and tone only.

— RoutineKit

The “Quiet clients” Notion view that saves your Monday

RoutineKit — Mon, 24 Aug 2026 21:07:25 +0000

The “Quiet clients” Notion view that saves your Monday

Freelance CRM usually dies the same way.

You have a Clients database. You have notes. You have “next actions” you swore you’d update.

Then Monday arrives and you can’t answer a simple question:

Who went quiet — and what should I do about them this week?

Not “every contact.”

Quiet clients. The ones who ghosted mid-thread, delayed a yes, or vanished after a proposal.

Build one saved view (not a bigger CRM)

If you already track clients in Notion, add (or reuse) these properties:

Property	Why
Stage	Lead / Active / Waiting / Closed
Last contact	Date
Next action	One verb phrase
Quiet?	Checkbox or a formula/view filter

Then create a saved view named Quiet clients:

Filter: Stage is Waiting or Last contact older than N days (7–14 works for most solos)
Sort: Last contact ascending (oldest silence first)
Show: Name · Next action · Last contact only

That’s it. No pipeline theater.

Monday ritual (10 minutes)

Open Quiet clients
Pick three names max
For each: one next action (bump email / close-lost / book call)
Write the action in the row before you open ChatGPT
Send or schedule — then leave the rest alone until next Monday

AI helps after the list exists.

Without the view, the model just helps you ramble into more silence.

Soft CTA

Want a full weekly Notion ops board (Tasks · Content · Clients · Money) with AI prompt slots and a rebuild guide?

→ https://routinekit.gumroad.com/l/vnlmtc ($9)

Prefer a free taste first (Tasks schema + 3 prompts):

→ https://routinekit.gumroad.com/l/alxurc

— RoutineKit

You use ChatGPT. Why does one proposal still eat your whole day?

RoutineKit — Mon, 24 Aug 2026 01:30:35 +0000

If you sell your own work, you already know the pattern.

A lead replies. You open ChatGPT.

Three hours later you still have a mushy draft, a half-written quote, and a calendar that quietly lost the afternoon.

The model isn’t the bottleneck.

The blank-page ritual is.

What actually burns the day

Most solo founders don’t lose the day “writing.”

They lose it to restarting:

Re-explaining the offer to the model every time
No fixed order (outline → draft → tighten → risk reversal → send)
Quotes invented in chat with fuzzy scope
No checklist, so you polish forever instead of shipping

More prompts won’t fix a missing weekly slot.

A lighter operating move

Treat proposals and quotes like appointments, not inspiration:

One slot on the calendar (45–90 minutes max)
One SOP order you reuse every time
Structured prompts (role · constraints · output format)
A filled example so you know what “sendable” looks like
A checklist that ends the session

Same week, same ritual. Only the client inputs change.

Soft CTA

If you want a free taste of the “structured prompt + checklist” style, start here (no purchase):

→ https://routinekit.gumroad.com/l/alxurc

If you want the full weekly operating system (proposals, content, replies, quotes, weekly review) as a $17 markdown playbook:

→ https://routinekit.gumroad.com/l/qkvnpp

— RoutineKit

Maker here — if your proposals still balloon past two hours, tell me where you get stuck (outline, pricing, or tone) and I’ll reply with a tighter sequence.

Pretty Notion templates won't run your week for you

RoutineKit — Sat, 22 Aug 2026 12:10:50 +0000

If your Notion looks beautiful and your week still feels scattered, the problem usually isn't "more pages."

It's that tasks, content ideas, client follow-ups, and cash notes live in four different places — so every Monday you reopen the same question:

What am I actually looking at this week?

ChatGPT is open in another tab. That doesn't help if nothing tells you which prompt belongs to which row.
Content: Name · Channel · Status (Idea → Hook locked → Draft → Edit → Scheduled → Published) · Publish date · Hook · CTA · Proof · URL.

Proof is required on purpose: a real experience, number, or case for every piece. Cheapest anti-hallucination guard there is, because you fill it before asking an LLM to draft anything.

Clients: Name · Stage (Lead / Meeting / Proposal / Quote / Contract / Active / Collecting / Closed) · Next action · Next date · Amount hint · Temperature. Next action is a verb with an implied deadline — "Thu send quote," not "follow up."

Money: Name · Type (In / Out / Expected in / Expected out) · Amount · Date · Category · Proof. Weekly cash sense only — not tax filing, not a replacement for real books.

Rebuild in ~20 minutes (no live duplicate link)

A lot of "Notion templates" assume a one-click duplicate.

That fails when you want the structure inside your own workspace, under your own permissions, without inheriting someone else's clutter.

So the practical approach is:

Create a root page: Weekly Ops Board
Add the four databases with explicit properties
Add weekly views (This Week Board / Publish Pipeline / Touch this week / This Week Ca ## A simple weekly rhythm

Monday: 25m board cleanup + pick today's 3
Wednesday: 40m content slots
Friday: 20m light cash sketch
Sunday: 25m review, lock next week

110 minutes. No second brain museum. No 40 linked databases.

What a wired prompt looks like

Role: Weekly ops coach for a solo operator. Goal: from the Inbox/Doing list below, pick exactly 3 tasks to finish today.

Rules: select exactly 3, label everything else defer / delegate / delete, prefer P0 and near deadlines and revenue-critical work, and keep the sum of est. minutes at or under 80% of available time.

Output: today's 3 with est. min and a one-line why, a defer list with reasons, and which rows to flip to "Today" in Notion.

Two details make it usable: the 80% capacity rule (without it the model hands you a 9-hour day), and that last output line — every AI turn has to end as a row edit inside the board, not a paragraph in a chat window you'll never reopen.

Also: replace only the placeholders. Rewriting the Rules section is how you get mush. And check facts, money, dates, and names before anything goes out — AI output is a draft.

Soft CTA

If you want the full Markdown schema + prompt slots + a filled example week as a downloadable pack, I published it as Solo Weekly Notion Ops Board + AI Prompts ($9):

https://routinekit.gumroad.com/l/vnlmtc

Prefer to try before buying? Free sample (Tasks schema + 3 prompts):

https://routinekit.gumroad.com/l/alxurc

Unofficial template — not affiliated with or endorsed by Notion Labs, Inc. Not financial or tax advice.

— RoutineKit

I'm the maker. Questions about property choices or "how do you avoid Notion bloat?" welcome in the comments.
sh)

Paste a small prompt library (pick-today's-3, stuck-task breakup, content batch, Friday cash sketch, Sunday review)
Replace example rows with your real week

The dashboard page is just a callout (this week's 3 goals + a do-not list) plus those four linked views. One scroll shows tasks, people, content, money. That's the whole point.

A minimum weekly schema (not an aesthetic dashboard)

For solo founders and freelancers, I've found four databases are enough:

DB	Job
Tasks	What ships this week
Content	Publish pipeline
Clients	Next action, not a full CRM
Money	Cash in / cash out sketch

Two shared properties do most of the work:

Week — Select: Last week / This week / Next week
AI Prompt — Text: which prompt slot you used on this row (T-1, C-2, ...)

That Week select is the trick. Every dashboard view filters on Week = This week, so the board empties and refills itself instead of growing into a graveyard. And the prompt slot means the model isn't a random chat — it's wired to the board.

Skip relations on day one. You can add them in week three if you still want them. (You usually don't.)

The properties that actually earn their keep

Tasks: Name (start with a verb — "Send quote") · Status (Inbox / Today / Doing / Waiting / Done / Canceled) · Priority (P0-P2) · Due · Area (Sales / Delivery / Content / Ops / Learning) · Est. min · Actual min.

Waiting separates "I'm blocked on me" from "I'm blocked on someone else's reply" — completely different problems on a Friday.

Stop Quoting "Around $500": A 5-Part Freelance Ops Loop You Can Run With AI

RoutineKit — Sat, 22 Aug 2026 11:47:24 +0000

Most freelance money leaks don't happen in the work. They happen in the paperwork loop around it: the quote you typed into a chat window, the "tiny tweak" you never priced, the receipts living in your camera roll until tax week.

Here's the loop I run, plus the prompts I paste into ChatGPT/Claude at each step. Not legal or tax advice: these produce drafts, and a human still checks before you sign or file.

1. Quote: itemize or lose the argument

"Around $500" is not a quote. It's a future dispute. A usable quote has line items, exclusions, a revision count, and a validity window.

Write a client quote.
Project: {{one-line scope}} | Deliverables: {{list}}
Rate basis: {{hourly/day/fixed + number}}
Output: (1) itemized table, hours and price per line,
(2) "Included / Not included" two-column list,
(3) revision policy: {{N}} rounds, extras priced,
(4) validity: {{14}} days.
No filler adjectives. Client must approve or reject each line.

Before: "Website redesign - $2,000." After: 6 line items, 3 exclusions, 2 revisions, expires in 14 days. Discount requests now hit a line item instead of your self-esteem.

2. Light agreement

A $2k gig doesn't need 12 pages. It needs scope, payment schedule, IP transfer on final payment, kill fee, late-payment terms. Ask the model to draft it from the quote in under 700 words, and to flag in brackets anything you must confirm with a lawyer. That bracket rule is what stops it from inventing enforceable-sounding clauses.

4. Books: 10 minutes a week

One sheet: date, client, category, amount, tax-relevant y/n, note. Paste the week's receipts, let AI categorize, you correct. Ten minutes weekly beats a lost weekend in April.

5. Advisor prep

Before a CPA call, generate a one-pager: entity type, revenue range, top expense categories, three specific questions, three things you're unsure about. You'll get more out of 30 minutes than most people get from three meetings.

The pattern

Every step is the same shape: SOP, then a prompt with real variables, then a filled example, then a pre-send checklist. The variables do the work; empty ones give you generic mush you'd never send.

I packaged this loop out as templates:

Free sample (the full quote module): https://routinekit.gumroad.com/l/alxurc
Freelancer Quote, Contract & Tax Prep Kit, $15: https://routinekit.gumroad.com/l/opcrcu
Solo AI workflow kit, $12: https://routinekit.gumroad.com/l/lsbdil

Start with the free one. If the quote module doesn't change how your next proposal looks, the rest won't either.

Which step in your ops loop still eats a whole evening?

3. Change requests: a script, not a mood

"Can you just..." is where margin dies. One template: acknowledge, cite the scope line, then two options - swap for something in scope, or add it at {{price}} and +{{N}} days - then ask for a yes/no. Two options reframes it from "are you charging me?" to "which one?"

You Use ChatGPT. Your Workload Didn't Shrink. Here's a Weekly OS.

RoutineKit — Sat, 22 Aug 2026 10:34:48 +0000

You already use ChatGPT daily. Yet the week still fills up: proposals eat the evening, content stays at "I'll post this week," replies get uneven, quotes start from a blank page. More prompts don't fix that — cadence does. Here's the weekly OS I run solo.

Slot A — Client proposal (~45 min)

Give the model a spine: restate the client's problem in their words; the outcome in one measurable sentence; in scope / out of scope / assumptions; timeline by phase with a deliverable each; price + payment split. Prompt shape: role = senior consultant writing to a busy non-technical client; constraints = under 400 words, no jargon, one price, explicit out-of-scope list. The win isn't typing speed, it's never re-deciding the structure.

Slot B — Content batching (60-75 min weekly)

Dump 10 raw ideas from real work. Pick 3. Write each as claim + proof in one line. Let the model expand the middle; rewrite the first and last paragraph by hand. Then schedule. If it isn't claim + proof, it isn't a post yet.

Slot C — Inquiry replies + scope fences (30-40 min)

Good fit: confirm the problem, propose a 20-minute call, name your price range up front.
Unclear: ask exactly two questions, no more.
Not a fit: decline in three lines, refer someone.

Fence I reuse: "That's outside this phase — happy to quote it as a separate add-on."

Slot D — Quote + simple SOW (40-50 min)

Deliverables, revision count, change-order trigger, payment schedule, validity date. Template it once; only numbers change.

Slot E — Weekly review (25-35 min)

What shipped? What slipped, and why? What are next week's five slots? Lock the calendar before closing the laptop.

An LLM removes typing friction, not decision friction. Fixed slots remove the deciding.

The packaged version

Every slot as SOP + prompts + filled example + checklist: Solo Operator AI Weekly System — $17
Narrower pieces: AI Workflow Kit — $12 and Quote, Contract & Tax Prep Kit — $15 (not legal advice). I'm the maker; questions about running slot B or C on a real week are welcome in the comments.