Meerkat
●cover story · n° 01

Write theprompt,not a novel.

A small, opinionated prompt builder for people tired of pasting cargo-cult incantations from Reddit at 1am.

no signup
required
est. 2026
Start buildingSee the anatomy60s · in-browser · free in beta
◆specimen 01 — anatomy of a working promptfull dissection
i.who's writing?
Role.

the model gets a job,
not a personality.

ii.what's the job?
Task.

one verb, one outcome,
stated in plain English.

iii.for whom, how long?
Context.

who reads it, how long,
what tone it lives in.

iv.what's off-limits?
Rules.

the lines you draw,
in writing, on purpose.

Sentinel for your sentences — published continuously since 2026eight modes · four questions · zero clichés
◐specimen 02 — the product, alive

Watch one get built.

Autoplay's on. Tap any tab to take the wheel — same flow you'll get inside the builder.

prompt.summarize.txt
61 tok·live
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
You are a careful editor.
Summarize the input below for: the executive team.
Length: an elevator pitch (2–3 sentences).
Angle: pain points + product opportunities.
 
Rules:
— Lead with the single most important point.
— Plain language. No filler, no hedging.
— Preserve concrete numbers, names, dates.
— Do not invent details that aren't in the input.
 
Input:
"""
{{transcript}}
"""
esc · close
autoplay·⌘C to copy
◐specimen 02b — the slot system

Curly braces are your variables.

Every prompt above uses the same trick. Type into a slot, watch the prompt fill in. That's the whole builder.

prompt.rewrite.txt
0/4 filled·live
01
02
03
04
05
06
07
08
09
10
11
12
You are a senior editor.
Rewrite the input below for: {{audience}}.
Tone: {{tone}}. Length: {{length}}.
 
Rules:
— Keep every concrete detail. Lose the throat-clearing.
— No clichés. Match how a real person talks.
 
Input:
"""
{{copy}}
"""

mustache primer

Anything in {{ }} is a slot.

Wrap a placeholder in double curly braces and Meerkat turns it into a labeled field. Type below — the prompt fills in real time.

how it gets injected

4 stages

  1. 01Template

    You write the prompt with {{slots}}.

  2. 02Detect

    Meerkat scans for slots and renders a field for each unique name.

  3. 03Fill

    You (or the next stage of an agent) supply values.

  4. 04Inject

    At copy/run time, every {{name}} is replaced with its value, then sent to the model.

Nothing is sent to a model until you copy or run. The template is the recipe; the values are the ingredients.

where slots belong

4 placements

  • ROLETop of the prompt, in the framing line. You are a {{role}}. Set the model's identity once, near the start.
  • PARAMSInline, mid-prompt, for short variables — audience, tone, length, format. Keep them on a single line so they read like dials, not paragraphs.
  • INPUTThe user's long content goes last, fenced inside triple """. This is where {{copy}}, {{sources}}, or {{paste}} live.
  • NEVERInside the rules section, mid-sentence. If the rule changes per request, hoist it into a {{...}} up top instead of burying it.

the rules

5 laws

  1. 01Two curly braces, no spaces. {{audience}} works. { audience } does not.
  2. 02Lowercase, snake_case names. {{user_request}} beats {{User Request}}. The builder uses the slug as the field key.
  3. 03Same name = same value. Use {{audience}} three times and you only fill it once.
  4. 04For long pasted input, fence it. Wrap {{copy}} in triple """ so the model knows where the user content starts and ends.
  5. 05Empty slots stay highlighted. Anything still in {{ }}at copy time is your reminder it's not done.
02 · specimen · bad prompt vs good
one of these gets shipped

vague. lazy. vibes.are out. Specifics ship.

Models don't need motivation, they need a job description. Meerkat asks the four questions a senior editor would ask before they ever touched the keyboard — then writes the prompt for you, in the format the model actually prefers.

exhibit a · the prompt that wastes 14 minutes

“hey can you summarize make this shorter this for me but make it good and professional and not too long please thanks”

no role · no audience · no rules→ AI-flavored mush
exhibit b · the prompt meerkat writes for you
ROLE
You are a careful editor.
TASK
Summarize the input below.
CONTEXT
Audience: execs.
Length: 2–3 sentences.
RULES
Lead with the point.
No filler. No clichés.
role · task · context · rules→ a paragraph that lands
§from the editor

The model isn't lazy. It's under-briefed. You wouldn't hand a contractor a Pinterest board and expect a kitchen.

the meerkat editorial board
we believe
specifics ship.
we don't
prime models.
we trade
ritual for rules.
we keep
the receipts.
◆field guide · the four bones, expanded

Get the bones right.The model does the rest.

Role, task, context, rules. Four ingredients, in that order, every time. The full dissection covers the smell tests, the rogues' gallery of common failures, and how to fix each one in under a minute.

Read the dissectionBuild one now
fig. 03 — on watch
04 · the main event · chat with meerkat

Just talk it out.

●headline · custom (chat with meerkat)

Describe what you want. We'll build the prompt together.

Live conversation. Prompt draft updates as you go. Paste an existing prompt and Meerkat audits it — what's lazy, what's vague, what to fix.

  • talk
    no forms. just say it.
  • draft
    prompt rewrites live.
  • audit
    paste, get a real critique.
Chat with Meerkatfree · 10 messages / session
chatting with meerkatlive

youI want to write better cold emails to founders.

meerkatCool. Fundraising, sales, intros, or recruiting? They each want a different prompt — saying “founder” is too thin.

draft v1 · updating
You are a sharp B2B copywriter. Write a 90-word cold email to {{founder}} at {{company}}…
fig. 04 — meerkat in conversationopen chat →
◇ or start from a template
See all templates →
≡01 / EDIT

Summarize

Long thing in, sharp thing out. Tweet to full memo, on demand.

4 questions · 60sbuild →
◊02 / READ

Analyze

Themes, risks, opportunities — cited from the source, not invented.

4 questions · 60sbuild →
✎03 / WRITE

Generate Copy

On-brand copy for a real channel. Three variations, no AI-flavored mush.

4 questions · 60sbuild →
{ }04 / PARSE

Extract Data

Messy text in, clean JSON out. No prose, no markdown fences, no excuses.

4 questions · 60sbuild →
↻05 / RESHAPE

Rewrite

Same idea, new audience or tone. Keeps every concrete detail you wrote.

4 questions · 60sbuild →
→06 / SCOPE

Plan

Goal in, ordered plan out. Dependencies and unknowns called out by name.

4 questions · 60sbuild →
⚐07 / REVIEW

Critique

Senior-editor pass on your work. Honest, specific, occasionally rude.

4 questions · 60sbuild →
≡01 / EDIT

Summarize

Long thing in, sharp thing out. Tweet to full memo, on demand.

4 questions · 60sbuild →
◊02 / READ

Analyze

Themes, risks, opportunities — cited from the source, not invented.

4 questions · 60sbuild →
✎03 / WRITE

Generate Copy

On-brand copy for a real channel. Three variations, no AI-flavored mush.

4 questions · 60sbuild →
{ }04 / PARSE

Extract Data

Messy text in, clean JSON out. No prose, no markdown fences, no excuses.

4 questions · 60sbuild →
↻05 / RESHAPE

Rewrite

Same idea, new audience or tone. Keeps every concrete detail you wrote.

4 questions · 60sbuild →
→06 / SCOPE

Plan

Goal in, ordered plan out. Dependencies and unknowns called out by name.

4 questions · 60sbuild →
⚐07 / REVIEW

Critique

Senior-editor pass on your work. Honest, specific, occasionally rude.

4 questions · 60sbuild →
05 · field guide · six approaches

One prompt style won't cut it.

Direct, structured, few-shot, chain-of-thought, just-in-time, sub-agent — six shapes of conversation, each one earning its keep on different tasks. Pick one in the builder and Meerkat reshapes the prompt around it.

Working in a sensitive domain — mental health, medical, legal, minors? Layer on Safeguardsand Meerkat injects non-negotiable policy rules the model can't reshape away.

Read the field guideTry in the builder
  1. 01 / 06→

    Direct

    One sentence in. One answer out. No ceremony.

    best for tasks the model already does wel…read →
  2. 02 / 06▦

    Structured

    Role · Task · Context · Rules. The Meerkat house style.

    best for most production work…read →
  3. 03 / 06❍❍

    Few-shot

    Show, don't tell. Two or three examples beat a paragraph of rules.

    best for output formats hard to describe …read →
  4. 04 / 06◐→●

    Chain-of-thought

    Make the model think out loud before it commits.

    best for multi-step reasoning, math, logi…read →
  5. 05 / 06⤓

    Just-in-time

    Don't stuff the context. Tell the model where to fetch what it needs.

    best for agents with tools (search, file …read →
  6. 06 / 06⌬

    Sub-agent

    One prompt to plan. Smaller prompts to execute. Less rope to hang yourself with.

    best for long, multi-step jobs (research,…read →
The Meerkat mascot running forward at full pace.
fig. 04 — exit, stage left
05 · end of line

Your next prompt should already be working.

Open the builder. Answer four questions. Walk away with a prompt you'd be embarrassed to ship without. Sixty seconds, tops — less than it took to read this paragraph.

Start buildingRead the anatomy