<?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: mari bostashvili</title>
    <description>The latest articles on DEV Community by mari bostashvili (@mari_bostashvili_65ce5346).</description>
    <link>https://dev.to/mari_bostashvili_65ce5346</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%2F4241780%2F631c983f-1461-4459-b42b-bd5ad4c87220.jpg</url>
      <title>DEV Community: mari bostashvili</title>
      <link>https://dev.to/mari_bostashvili_65ce5346</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/mari_bostashvili_65ce5346"/>
    <language>en</language>
    <item>
      <title>I benchmarked my repo-compression CLI on a 25k-star monorepo, and it found three bugs in itself</title>
      <dc:creator>mari bostashvili</dc:creator>
      <pubDate>Thu, 12 Nov 2026 21:11:12 +0000</pubDate>
      <link>https://dev.to/mari_bostashvili_65ce5346/i-benchmarked-my-repo-compression-cli-on-a-25k-star-monorepo-and-it-found-three-bugs-in-itself-2p5c</link>
      <guid>https://dev.to/mari_bostashvili_65ce5346/i-benchmarked-my-repo-compression-cli-on-a-25k-star-monorepo-and-it-found-three-bugs-in-itself-2p5c</guid>
      <description>&lt;p&gt;Every new Claude Code session on my projects started the same way. The agent re-read the repository, rediscovered the design tokens, and forgot what we decided yesterday. On a large repo that is a lot of tokens spent before any real work.&lt;/p&gt;

&lt;p&gt;So I built ai-ria, a small open-source CLI that writes a .ria/ folder into your project. Your agent reads that folder instead of the whole repo:&lt;/p&gt;

&lt;p&gt;an architecture summary, routes, components and conventions&lt;br&gt;
a compressed context pack, sized to a token budget&lt;br&gt;
memory that survives between sessions: decisions, warnings, design rules&lt;br&gt;
handoffs, so you can switch between Claude Code, Cursor and Codex without starting from zero&lt;br&gt;
design tokens, read from your CSS or from Figma exports&lt;br&gt;
It makes no LLM calls. It is static analysis, so it's deterministic, needs no API key and runs in seconds.&lt;/p&gt;

&lt;p&gt;npx ai-ria analyze .&lt;br&gt;
The benchmark&lt;br&gt;
Before announcing it, I ran the published version on two well-known Next.js repositories.&lt;/p&gt;

&lt;p&gt;vercel/commerce (14k stars) worked as expected. Roughly 30–42k tokens of source became a ~7.6k-token pack, with 12 routes and 31 components detected.&lt;/p&gt;

&lt;p&gt;dubinc/dub (25k stars, a Turborepo monorepo) went worse. From the repo root, ai-ria reported:&lt;/p&gt;

&lt;p&gt;framework:          Node&lt;br&gt;
routes:             0&lt;br&gt;
components:         2&lt;br&gt;
security findings:  382 (2 critical/high)&lt;br&gt;
All three numbers were wrong:&lt;/p&gt;

&lt;p&gt;The monorepo was read by its tooling root. The root package.json only lists tooling. The real app lives in apps/web, which has 456 routes.&lt;br&gt;
Components outside components/ were invisible. dub keeps hundreds of kebab-case components in ui/ folders.&lt;br&gt;
The security report was almost all noise. 379 of the 380 "insecure http URL" findings were inline SVG namespaces (xmlns="&lt;a href="http://www.w3.org/2000/svg%22" rel="noopener noreferrer"&gt;http://www.w3.org/2000/svg"&lt;/a&gt;). The two "critical" findings were a local Docker URL in a hint message and a password in a Playwright test.&lt;br&gt;
The fixes (0.2.1)&lt;br&gt;
Workspace roots (npm/yarn/bun workspaces, pnpm, Turborepo, Lerna, Nx) are detected, each package is analyzed on its own, and the repo is described by its main app.&lt;br&gt;
Components count from ui/, widgets/, primitives/ and elements/ folders as well. Stories, tests and fixtures don't count.&lt;br&gt;
XML namespaces and localhost URLs are no longer flagged, and e2e suites (playwright/, cypress/, e2e/) are treated as tests. Remote http:// URLs and remote credentials are still flagged.&lt;br&gt;
Same command on dub after the fix:&lt;/p&gt;

&lt;p&gt;framework:          Next.js (apps/web, 9 workspaces)&lt;br&gt;
routes:             456&lt;br&gt;
components:         846&lt;br&gt;
security findings:  0 (0 critical/high)&lt;br&gt;
context pack:       ~13504 tokens (raw ~1989768)&lt;br&gt;
That's about 2M tokens of source down to a ~13.5k-token pack.&lt;/p&gt;

&lt;p&gt;What I learned&lt;br&gt;
Running your tool on someone else's real repo, before anyone else does, is the cheapest code review you'll get. My own test fixtures were all single-package apps with a components/ folder, so the tests stayed green while the tool was wrong on a whole class of real projects.&lt;/p&gt;

&lt;p&gt;Try it&lt;br&gt;
npx ai-ria analyze .                 # writes .ria/&lt;br&gt;
npx ai-ria agent-pack .              # AGENT_PACK.md, the file the next agent reads&lt;br&gt;
claude mcp add ai-ria -- npx -y ai-ria mcp   # or use it as an MCP server in Claude Code&lt;br&gt;
GitHub: &lt;a href="https://github.com/maribostashvili-93/ai-ria" rel="noopener noreferrer"&gt;https://github.com/maribostashvili-93/ai-ria&lt;/a&gt; (MIT, Node ≥ 20)&lt;/p&gt;

&lt;p&gt;It's early. I'd like to know where it breaks on your repo, and what you use today to give agents context between sessions.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>webdev</category>
      <category>productivity</category>
      <category>javascript</category>
    </item>
  </channel>
</rss>
