<?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: Shaolin Zhou</title>
    <description>The latest articles on DEV Community by Shaolin Zhou (@shaolin_zhou_cbc100b2d8fd).</description>
    <link>https://dev.to/shaolin_zhou_cbc100b2d8fd</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%2F4247794%2F3450a0a6-114c-48a8-8891-e27d3ef4113a.png</url>
      <title>DEV Community: Shaolin Zhou</title>
      <link>https://dev.to/shaolin_zhou_cbc100b2d8fd</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/shaolin_zhou_cbc100b2d8fd"/>
    <language>en</language>
    <item>
      <title>Vibe Coding made code cheap. It didn't make "done" trustworthy.</title>
      <dc:creator>Shaolin Zhou</dc:creator>
      <pubDate>Mon, 16 Nov 2026 12:56:20 +0000</pubDate>
      <link>https://dev.to/shaolin_zhou_cbc100b2d8fd/vibe-coding-made-code-cheap-it-didnt-make-done-trustworthy-b97</link>
      <guid>https://dev.to/shaolin_zhou_cbc100b2d8fd/vibe-coding-made-code-cheap-it-didnt-make-done-trustworthy-b97</guid>
      <description>&lt;p&gt;Vibe Coding made code cheap. It didn't make "done" trustworthy.&lt;/p&gt;

&lt;p&gt;When AI agents write most of your code, "it's finished" becomes a claim you have to verify. I ran into this while building a multi-physics simulation engine with AI coding agents: I had no reliable way to see which "done" features actually had verification evidence behind them.&lt;/p&gt;

&lt;p&gt;Tasks lived in chat context that scrolled away. A new idea mid-session had nowhere to land. And nothing connected a feature to the test or benchmark that was supposed to prove it works.&lt;/p&gt;

&lt;p&gt;That is a project management problem, and engineering projects solved it long ago with task hierarchies, state machines, and traceable verification. So I applied that discipline to agent-driven development and built &lt;strong&gt;Lite ADLM&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it is
&lt;/h2&gt;

&lt;p&gt;Lite ADLM is a local-first task tree that humans and AI agents work on together:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Plain JSON files on disk.&lt;/strong&gt; No database.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A pure-Python CLI&lt;/strong&gt; that uses only the standard library, so there is nothing to install.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;An optional real-time board&lt;/strong&gt; (Node.js + Socket.IO), so you and the agent see the same data. You can add or reprioritize tasks while the agent works, and it sees them on its next query.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Five lists&lt;/strong&gt;: architecture, features, tests, benchmarks, bugs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Relations between items&lt;/strong&gt;: &lt;code&gt;depends-on&lt;/code&gt;, &lt;code&gt;verifies&lt;/code&gt;, &lt;code&gt;blocks&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A &lt;code&gt;gaps&lt;/code&gt; command&lt;/strong&gt; that audits which features have no recorded verification.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fr07ecg7rphhzgeoi2sij.gif" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fr07ecg7rphhzgeoi2sij.gif" alt="Agent writes tasks, board updates in real time" width="760" height="487"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it in 30 seconds
&lt;/h2&gt;

&lt;p&gt;You need Python 3 (I tested on 3.12). Nothing else.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/alainpdeai/lite-adlm-en.git lite-adlm
&lt;span class="nb"&gt;cd&lt;/span&gt; lite-adlm
python scripts/litetask_cli.py init
python scripts/litetask_cli.py add --category func --title &lt;span class="s2"&gt;"First Task"&lt;/span&gt; --agent &lt;span class="s2"&gt;"AI Agent"&lt;/span&gt;
python scripts/litetask_cli.py list --category func
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;[ ] First Task  P=Medium  S=open
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Now ask the audit what's missing:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python scripts/litetask_cli.py gaps
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;GAP Feature not verified by test: 1 item(s)  (owner=func referrer=test relation=verifies)
   - [func] First Task
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;To close the gap, add a test item and link it to the feature with a &lt;code&gt;verifies&lt;/code&gt; relation:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python scripts/litetask_cli.py add --category &lt;span class="nb"&gt;test&lt;/span&gt; --title &lt;span class="s2"&gt;"Verify First Task"&lt;/span&gt; --agent &lt;span class="s2"&gt;"AI Agent"&lt;/span&gt;
python scripts/litetask_cli.py link --category &lt;span class="nb"&gt;test&lt;/span&gt; --title-or-id &lt;span class="s2"&gt;"Verify First Task"&lt;/span&gt; \
  --relates func:&lt;span class="o"&gt;&amp;lt;&lt;/span&gt;func_id&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;:verifies
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Run &lt;code&gt;gaps&lt;/code&gt; again and that feature no longer appears.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the audit found on my own project
&lt;/h2&gt;

&lt;p&gt;I tracked 220+ task items (including subtasks and archived ones) while building the simulation engine. When I ran &lt;code&gt;gaps&lt;/code&gt; on the active task tree, it listed &lt;strong&gt;69 items&lt;/strong&gt; (including subtasks) with no recorded verification link. &lt;strong&gt;37 of them were already marked done.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;That does not mean the code was untested. It means the evidence wasn't traceable: nothing in the task graph connected those features to a test or benchmark. Before the audit, that was a hidden assumption. After it, it was a visible backlog.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it doesn't do (honest limitations)
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The verification rule is advisory, not enforced.&lt;/strong&gt; A rule in the skill file tells the agent not to close a feature until a test or benchmark links to it. The tool itself allows &lt;code&gt;open → done&lt;/code&gt; directly. Illegal jumps such as &lt;code&gt;open → verified&lt;/code&gt; are rejected, but the "no verification, no done" part relies on the agent following the rule, and &lt;code&gt;gaps&lt;/code&gt; audits it afterwards. That is exactly why the audit command exists.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It uses a local trust model.&lt;/strong&gt; The server binds to &lt;code&gt;127.0.0.1&lt;/code&gt; by default, and the data API trusts caller-supplied identity. Don't expose it to an untrusted network.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It's not a Jira or GitHub Issues replacement.&lt;/strong&gt; It's for one developer (or a small group) working locally with agents, not for enterprise permissions or managed cloud collaboration.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;I've only tested it with Claude Code.&lt;/strong&gt; The Cursor and Aider configs in the README are example templates I haven't verified.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Design choices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Files over a database&lt;/strong&gt;, so the task tree is something you can read, diff, and back up.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Standard library only&lt;/strong&gt; for the CLI, so it works anywhere Python does.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optimistic concurrency on the HTTP API&lt;/strong&gt; (&lt;code&gt;baseRev&lt;/code&gt; and &lt;code&gt;409 Conflict&lt;/code&gt;) to prevent lost updates, and atomic temp-file replacement in CLI mode to prevent corrupted files.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A test suite you can run yourself.&lt;/strong&gt; There are 13 fast graders; with &lt;code&gt;init&lt;/code&gt; run and the server started, they pass 13/13 on my machine.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Feedback welcome
&lt;/h2&gt;

&lt;p&gt;If you run long sessions with Claude Code, Cursor, or any agent that can run shell commands, I'd love to hear what feels clunky, especially around connecting your agent and the &lt;code&gt;verifies&lt;/code&gt; workflow.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Repo (Apache 2.0): &lt;a href="https://github.com/alainpdeai/lite-adlm-en" rel="noopener noreferrer"&gt;https://github.com/alainpdeai/lite-adlm-en&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Quick Start, agent setup, and docs are all in the README.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>agents</category>
      <category>ai</category>
      <category>programming</category>
      <category>softwareengineering</category>
    </item>
  </channel>
</rss>
