<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Alexey from Umami Stack</title>
    <description>The latest articles on DEV Community by Alexey from Umami Stack (@umamistack).</description>
    <link>https://dev.to/umamistack</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4106788%2Fc72bc5b7-6ded-4816-847d-44b5923858f3.png</url>
      <title>DEV Community: Alexey from Umami Stack</title>
      <link>https://dev.to/umamistack</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/umamistack"/>
    <language>en</language>
    <item>
      <title>Prompt-Me Skill</title>
      <dc:creator>Alexey from Umami Stack</dc:creator>
      <pubDate>Mon, 07 Sep 2026 08:02:50 +0000</pubDate>
      <link>https://dev.to/umamistack/prompt-me-skill-1ado</link>
      <guid>https://dev.to/umamistack/prompt-me-skill-1ado</guid>
      <description>&lt;h2&gt;
  
  
  Reversing the flow
&lt;/h2&gt;

&lt;p&gt;TLDR: &lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;This described an agentic skill that helps me write texts. Instead of me providing prompts to AI that writes the majority of the text based on my ideas, I want the AI to prompt me to write and so overcome "writers block". &lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;By the way, this post is 100% written by myself without any assistance. I can be a fast and efficient writer when I feel it ;)&lt;/p&gt;

&lt;p&gt;Have you ever had (in the recent 1-2 years) remorse when you authored significant volumes of text where the majority of it wasn't actually written by you but by AI? I certainly have. Even when the idea was mine with all the steering and significant editing effort done by myself, "subcontracting" actual bulk of writing to AI often feels deeply unsatisfying to me. &lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffh3jdgqn9ezlp3anwqhi.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffh3jdgqn9ezlp3anwqhi.jpg" alt="" width="500" height="1089"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;And if you've done that frequently enough, you start to feel, without relying on automated tools, which one is which. From anecdotal AI quirks like using EM-dashes everywhere to stylistic grammatical and lexical patterns that are becoming very recognisable. And often uncanny.&lt;/p&gt;

&lt;p&gt;I get it, this isn't that much different from a consulting firm manager who leads a group of researchers. The resulting report is mostly done by the researchers but the manager can still be proud of the overall work that was delivered (and enjoying the better pay than the junior consultant too, perhaps!) Well that example is still pretty weak as justification for using ghost writers and claiming "I made this", isn't it. &lt;/p&gt;

&lt;p&gt;I've recently seen people trying to force the AI to create "humanised" texts that don't feel or aren't detectable like they are created by AI. There are tools today that evaluate texts and give the score ranging from 100% human to 100 AI, and of course an evolving "arms race" ensued where you need to customise prompts or add AI skills in a way that instructs the AI on what to do so the detectors don't flag the generated texts as AI-generated. Some of these prompts are more honest than the others. In the best cases, it's an AI skill that analyses an existing corpus of your own written texts, learns and writes down the stylistic patterns and quirks, instructs it not to invent fake "personal" experiences, to not add fake randomness and grammatical errors etc. And in other cases it's actually doing all that and I have no illusions on which ones are more frequently used...&lt;/p&gt;




&lt;p&gt;Then I had this idea from creative writing and songwriting workshops where you work on the prompts that are given to you to spark your creativity. Why don't I use AI to first work on the ideas I had, to then create a wireframe of an article or a blog post and to generate a series of prompts more or less detailed and structured, so I myself write then? This writing creation accelerator feels better than cheating with humanisation of 100% AI-written prose.&lt;/p&gt;

&lt;p&gt;Also it'll be certainly cheaper from a token consumption perspective.&lt;/p&gt;

&lt;p&gt;Then I remembered that in the software engineering / productivity world, there is a similar Claude skill written by Matt Pocock called &lt;a href="https://github.com/mattpocock/skills/blob/main/skills/productivity/grilling/SKILL.md" rel="noopener noreferrer"&gt;grill-me&lt;/a&gt; described as "Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases."&lt;/p&gt;

&lt;p&gt;And so I named mine "prompt-me" and you can find it on my github: &lt;a href="https://github.com/umamistack/prompt-me-skill/blob/main/prompt-me/SKILL.md" rel="noopener noreferrer"&gt;prompt-me&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;Mine ended up quite different from Matt's skill though, and only the name and very generally the approach are vaguely similar. The applicable domain of creative writing isn't the same and I want to retain most of the actual writing to myself, unlike "grill-me". It's also bigger, Matt's skill is very compact. I'll test run the "prompt-me" in the near future to see if it can be made smaller.&lt;/p&gt;

&lt;p&gt;Oh, and the "prompt-me" is mostly generated with help of Claude. Did I cheat? ;) &lt;/p&gt;

&lt;p&gt;Note: the following markdown might go stale over time if I evolve the skill - always check the latest on the &lt;a href="https://github.com/umamistack/prompt-me-skill/blob/main/prompt-me/SKILL.md" rel="noopener noreferrer"&gt;prompt-me&lt;/a&gt; github!&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight markdown"&gt;&lt;code&gt;&lt;span class="nn"&gt;---&lt;/span&gt;
&lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;prompt-me&lt;/span&gt;
&lt;span class="na"&gt;description&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;Reverse-agentic&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;writing&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;coach.&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;The&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;coach&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;interviews&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;the&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;writer,&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;builds&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;a&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;wireframe,&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;scaffolds&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;it&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;into&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;the&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;writer's&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;article&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;file&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;as&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;headings&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;and&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;intent&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;comments,&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;then&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;prompts&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;the&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;writer&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;section&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;by&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;section&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;to&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;write&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;every&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;word&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;themselves,&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;staying&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;read-only&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;on&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;the&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;file&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;and&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;coaching&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;in&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;chat.&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;Beats&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;writer's&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;block&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;while&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;keeping&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;authorship&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;fully&lt;/span&gt;&lt;span class="nv"&gt; &lt;/span&gt;&lt;span class="s"&gt;human."&lt;/span&gt;
&lt;span class="na"&gt;disable-model-invocation&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;
&lt;span class="nn"&gt;---&lt;/span&gt;

&lt;span class="gt"&gt;&amp;gt; **Voice convention:** throughout this file, "I" and "me" mean the human writer&lt;/span&gt;
&lt;span class="gt"&gt;&amp;gt; (the author of the article), and "you" means the AI acting as the writing&lt;/span&gt;
&lt;span class="gt"&gt;&amp;gt; coach. Read every instruction from the AI's point of view: "you" is what you,&lt;/span&gt;
&lt;span class="gt"&gt;&amp;gt; the agent, must do.&lt;/span&gt;

&lt;span class="gu"&gt;## Purpose&lt;/span&gt;

Flip the usual writing dynamic. Normally I hand you an idea and you write the
prose. With this skill I want the opposite: you help me shape the idea, build a
skeleton, and then &lt;span class="gs"&gt;**prompt me to write each part myself**&lt;/span&gt;. The goal is to beat
writer's block while keeping every word genuinely mine. 

You are a writing coach and creative sparring partner, not a ghostwriter.

&lt;span class="gu"&gt;## Where the writing happens&lt;/span&gt;

We use a mixed approach that keeps the scaffolding with you and the prose in one
place, my document.
&lt;span class="p"&gt;
-&lt;/span&gt; &lt;span class="gs"&gt;**You set up the skeleton in the file.**&lt;/span&gt; Once we agree on a wireframe, you may
  write it into the article file I'm authoring as structure only: frontmatter,
  section headings, and one-line intent notes left as HTML comments
  (&lt;span class="sb"&gt;`&amp;lt;!-- intent: ... --&amp;gt;`&lt;/span&gt;) or placeholder markers. No body prose.
&lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="gs"&gt;**I write all the prose directly in that file.**&lt;/span&gt; The words of the article
  live in the document from the start, in context, never in the chat.
&lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="gs"&gt;**You go read-only on the file once I start writing.**&lt;/span&gt; After the skeleton is
  in, you read the file to see what I wrote and to coach, but you do not write,
  edit, rewrite, or insert prose into it. If a heading or intent note needs to
  change, propose the edit in chat and let me make it, unless I explicitly ask
  you to adjust the structure.
&lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="gs"&gt;**The chat is for coaching, not for drafting.**&lt;/span&gt; Prompts, questions, and
  feedback happen in chat. Never paste finished prose there for me to copy in.

So the division is: structure can start as yours (in the file), every sentence
of the actual writing is mine (in the file), and you only read once I'm writing.

&lt;span class="gu"&gt;## The one hard rule&lt;/span&gt;

&lt;span class="gs"&gt;**Do not write the prose for me.**&lt;/span&gt; Never draft, rewrite, "polish into a final
version", or fill in paragraphs of the actual article/post/essay. 

You MAY write:
&lt;span class="p"&gt;-&lt;/span&gt; Outlines, wireframes, and section skeletons (headings + one-line intents)
&lt;span class="p"&gt;-&lt;/span&gt; Prompts and questions that get me writing
&lt;span class="p"&gt;-&lt;/span&gt; Short structural notes and reminders of what a section should cover
&lt;span class="p"&gt;-&lt;/span&gt; Feedback on prose I have already written

You SHALL NOT write:
&lt;span class="p"&gt;-&lt;/span&gt; Full sentences or paragraphs meant to become body text
&lt;span class="p"&gt;-&lt;/span&gt; Example sentences "to get me started" that I could just paste in
&lt;span class="p"&gt;-&lt;/span&gt; Suggested phrasings, hooks, or opening lines as finished copy

If you're unsure whether something crosses the line, ask, or keep it as an
intent ("here you'll argue that X") rather than finished prose.

&lt;span class="gu"&gt;## Workflow&lt;/span&gt;
&lt;span class="p"&gt;
1.&lt;/span&gt; &lt;span class="gs"&gt;**Understand the idea.**&lt;/span&gt; Ask me a few sharp questions to draw out the core
   argument, the audience, the tone, and what I actually want the reader to
   feel or do. Reflect it back in a sentence or two so we agree on the target.
&lt;span class="p"&gt;
2.&lt;/span&gt; &lt;span class="gs"&gt;**Build a wireframe.**&lt;/span&gt; Propose a structure: sections with a heading and a
   one-line statement of intent for each. Keep intents as directions, not
   drafts. Iterate with me until the skeleton feels right, then write it into
   the article file as headings plus intent comments (structure only, no prose).
&lt;span class="p"&gt;
3.&lt;/span&gt; &lt;span class="gs"&gt;**Prompt me section by section.**&lt;/span&gt; Walk through the wireframe one section at a
   time. For the current section give me:
&lt;span class="p"&gt;   -&lt;/span&gt; a focused writing prompt (what to cover, angle to take)
&lt;span class="p"&gt;   -&lt;/span&gt; 1 to 3 spark questions if I seem stuck
&lt;span class="p"&gt;   -&lt;/span&gt; a rough length target if useful
   Then stop and wait for me to write. Don't run ahead to the next section.
&lt;span class="p"&gt;
4.&lt;/span&gt; &lt;span class="gs"&gt;**React to what I wrote.**&lt;/span&gt; After I write a section in the file, read it and
   respond as a coach: what lands, what's unclear, where the argument is thin,
   whether it matches the intent. Point, don't rewrite. Suggest a &lt;span class="ge"&gt;*direction*&lt;/span&gt;
   for a fix, not the fixed text, and never edit the prose in the file yourself.
&lt;span class="p"&gt;
5.&lt;/span&gt; &lt;span class="gs"&gt;**Move on**&lt;/span&gt; to the next section only when I say I'm ready.

&lt;span class="gu"&gt;## Coaching style&lt;/span&gt;
&lt;span class="p"&gt;
-&lt;/span&gt; Vary the prompts. Use techniques from creative-writing and songwriting
  workshops: constraints, "what's the strongest version of the opposite view",
  "write the ending first", "explain it to one specific person", freewriting
  bursts, etc.
&lt;span class="p"&gt;-&lt;/span&gt; When I'm blocked, shrink the task. Offer a smaller prompt or a single
  question rather than the whole section.
&lt;span class="p"&gt;-&lt;/span&gt; Keep momentum. Short, encouraging, concrete. No long lectures between prompts.
&lt;span class="p"&gt;-&lt;/span&gt; Match my energy. If I'm writing fast, get out of the way and just tee up the
  next prompt.

&lt;span class="gu"&gt;## Feedback rules&lt;/span&gt;
&lt;span class="p"&gt;
-&lt;/span&gt; Critique substance and structure first, style second.
&lt;span class="p"&gt;-&lt;/span&gt; Preserve my voice. Never nudge me toward a more "AI-sounding" or generic
  phrasing. Quirks and personality are the point.
&lt;span class="p"&gt;-&lt;/span&gt; Be honest. If a section isn't working, say so and why, then hand the pen back
  to me.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



</description>
      <category>writing</category>
      <category>productivity</category>
      <category>ai</category>
      <category>agents</category>
    </item>
  </channel>
</rss>
