<?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: Kontenlaris Studio</title>
    <description>The latest articles on DEV Community by Kontenlaris Studio (@kontenlarisstudio).</description>
    <link>https://dev.to/kontenlarisstudio</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%2F4257164%2F76e299fc-a5d9-460b-a0b8-cd17edb913f0.png</url>
      <title>DEV Community: Kontenlaris Studio</title>
      <link>https://dev.to/kontenlarisstudio</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/kontenlarisstudio"/>
    <language>en</language>
    <item>
      <title>Building CrewKit: a skeleton that grew two apps</title>
      <dc:creator>Kontenlaris Studio</dc:creator>
      <pubDate>Sat, 21 Nov 2026 10:45:00 +0000</pubDate>
      <link>https://dev.to/kontenlarisstudio/building-crewkit-a-skeleton-that-grew-two-apps-1fgm</link>
      <guid>https://dev.to/kontenlarisstudio/building-crewkit-a-skeleton-that-grew-two-apps-1fgm</guid>
      <description>&lt;p&gt;&lt;em&gt;This post is written as part of the &lt;a href="https://kiroween.devpost.com" rel="noopener noreferrer"&gt;Kiroween 2026&lt;/a&gt; hackathon — it documents how I built my Skeleton Crew entry, &lt;a href="https://devpost.com/software/crewkit" rel="noopener noreferrer"&gt;CrewKit&lt;/a&gt;, with Kiro.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The itch
&lt;/h2&gt;

&lt;p&gt;Every automation project I start begins the same way: a scheduler, config loading, state that survives a crash, and a way to notify a human when something happens. I have written those four things from scratch more times than I can count, and every framework that offers them brings thousands of lines and a dependency tree along for the ride.&lt;/p&gt;

&lt;p&gt;So for the Skeleton Crew category — &lt;em&gt;a lean skeleton template, proven by two distinct applications&lt;/em&gt; — I built the bones exactly once, in the open, with an AI agent doing most of the typing under unusually strict house rules.&lt;/p&gt;

&lt;p&gt;The result: &lt;strong&gt;CrewKit&lt;/strong&gt;, a 433-line core in pure stdlib Python (zero dependencies), with two working apps on top of it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;a &lt;strong&gt;Telegram task-reminder bot&lt;/strong&gt; (&lt;code&gt;/add&lt;/code&gt;, &lt;code&gt;/list&lt;/code&gt;, &lt;code&gt;/done&lt;/code&gt;, due-task nudges), and&lt;/li&gt;
&lt;li&gt;a &lt;strong&gt;price-watch notifier&lt;/strong&gt; that extracts prices from JSON-LD and alerts when a target is hit.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Same bones. Two completely different creatures.&lt;/p&gt;

&lt;h2&gt;
  
  
  Spec first, code second
&lt;/h2&gt;

&lt;p&gt;The part of Kiro that changed how I work is not autocomplete — it is the &lt;code&gt;.kiro&lt;/code&gt; directory as a &lt;em&gt;contract&lt;/em&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Steering files&lt;/strong&gt; (&lt;code&gt;product.md&lt;/code&gt;, &lt;code&gt;tech.md&lt;/code&gt;, &lt;code&gt;structure.md&lt;/code&gt;) fixed the rules before any code existed: stdlib only, a 700-line budget for the core, apps may only touch the core's public API, every app must run offline in a &lt;code&gt;--demo&lt;/code&gt; mode.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A spec&lt;/strong&gt; (&lt;code&gt;requirements.md&lt;/code&gt; R1–R6, &lt;code&gt;design.md&lt;/code&gt;, &lt;code&gt;tasks.md&lt;/code&gt;) described the core module by module: config, clock, store, notify, runner — and the failure policy ("a raising job never kills the loop").&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A custom agent&lt;/strong&gt; (&lt;code&gt;crewkit-builder&lt;/code&gt;, created with &lt;code&gt;kiro-cli agent create&lt;/code&gt;) whose prompt pinned it to the steering + spec and ended with a hard rule: &lt;em&gt;you are not finished until &lt;code&gt;python3 tests/selftest.py&lt;/code&gt; passes.&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A hook&lt;/strong&gt; that re-runs the self-test whenever a Python file is edited, so a broken skeleton gets caught at edit time, not demo time.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Two Kiro CLI sessions later — one for the core, one for both apps — I had 32 passing tests, then 34, and I had barely written any of the implementation myself. My job had moved up a level: writing requirements sharp enough that "done" is checkable, then verifying the evidence instead of the prose.&lt;/p&gt;

&lt;h2&gt;
  
  
  My favorite thing about Kiro
&lt;/h2&gt;

&lt;p&gt;The finish line moved. In my old workflow, an AI session ended when the output &lt;em&gt;looked&lt;/em&gt; right. With specs + steering + a custom agent, the session ends when a deterministic test suite says it is right — the agent iterates against the tests, not against my patience. The core landed at 433 lines against its 700-line budget not because I counted, but because the steering file did.&lt;/p&gt;

&lt;p&gt;That is the change I am taking to every future project: &lt;strong&gt;give the agent a contract and a test, not a vibe.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Proof, not promises
&lt;/h2&gt;

&lt;p&gt;Everything in CrewKit is verifiable from a clean checkout, offline:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 tests/selftest.py                    &lt;span class="c"&gt;# 34 tests, ~0.25 s&lt;/span&gt;
python3 apps/telegram-task-bot/bot.py --demo &lt;span class="c"&gt;# prints a due-task reminder&lt;/span&gt;
python3 apps/price-watch/watch.py --demo     &lt;span class="c"&gt;# prints a price-drop alert&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;Repo (MIT, with the full &lt;code&gt;.kiro/&lt;/code&gt; directory): &lt;a href="https://github.com/kontenlaris-studio/crewkit" rel="noopener noreferrer"&gt;https://github.com/kontenlaris-studio/crewkit&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Live demo site: &lt;a href="https://kontenlaris-studio.github.io/crewkit/" rel="noopener noreferrer"&gt;https://kontenlaris-studio.github.io/crewkit/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Demo video (2:42): &lt;a href="https://www.youtube.com/watch?v=6Fuo2gpA0ok" rel="noopener noreferrer"&gt;https://www.youtube.com/watch?v=6Fuo2gpA0ok&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Devpost project: &lt;a href="https://devpost.com/software/crewkit" rel="noopener noreferrer"&gt;https://devpost.com/software/crewkit&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What I would do next
&lt;/h2&gt;

&lt;p&gt;More notifier sinks (ntfy, Slack), a third proof app (an RSS digest — the classic skeleton stress test), and a &lt;code&gt;crewkit new &amp;lt;app&amp;gt;&lt;/code&gt; scaffolder generated from the same spec workflow. The skeleton held up under two apps; a third will tell me if it is a template or just a coincidence.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Built with Kiro CLI (specs, steering, hooks, custom agents) for Kiroween 2026, Skeleton Crew category.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>kiro</category>
      <category>hackathon</category>
      <category>python</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
