Scroll, if you must ↓

Leaderboard

The 25 driest public repos

Public repositories shared on DRY.codes, ranked by originality. Internal is how little a repo repeats itself; Public is how original it is versus every other public repo. A wide Internal→Public gap means it overlaps other public code — forked, templated, or, increasingly, agent-generated. Click a column to re-rank, or hover a score for the breakdown.

Showing the top 25 of 45 public repositories, ranked by Public DRY (originality vs the whole corpus).

#Repository Internal Public Files Indexed
🥇 zhw2590582/ArtPlayer
91
82
135 2026-07-16
🥈 mukul975/Anthropic-Cybersecurity-Skills
88
70
120 2026-06-27
🥉 quotentiroler/aegis
90
70
38 2026-07-19
4 nousresearch/hermes-agent
87
67
141 2026-06-25
5 ZhuLinsen/daily_stock_analysis
88
67
139 2026-06-24
6 Jakubantalik/border-beam
92
67
26 2026-07-27
7 facebook/astryx
81
66
980 2026-07-29
8 koala73/worldmonitor
87
65
134 2026-06-25
9 NanmiCoder/MediaCrawler
82
65
132 2026-06-26
10 Max-Health-Inc/proxy-smart
84
64
826 2026-07-25
11 mauriceboe/TREK
85
64
140 2026-06-26
12 andrewyng/openworker
79
63
383 2026-07-27
13 quotentiroler/CactusRoute
89
63
12 2026-07-19
14 opendatalab/MinerU
79
62
145 2026-06-26
15 n0-computer/iroh
82
62
141 2026-06-27
16 aws/agent-toolkit-for-aws
80
60
145 2026-06-26
17 IceWhaleTech/CasaOS
86
60
141 2026-06-27
18 google-research/timesfm
79
60
81 2026-06-27
19 garrytan/gstack
84
60
69 2026-06-26
20 Serranegra/optifarm
81
60
38 2026-07-27
21 interviewstreet/hiring-agent
93
60
14 2026-06-27
22 microsoft/vscode
87
59
139 2026-06-25
23 shanraisshan/claude-code-best-practice
80
59
88 2026-06-26
24 every-app/open-seo
89
58
143 2026-06-26
25 Max-Health-Inc/prefab
83
57
196 2026-07-23

The problem

Duplication scales with your company

Add engineers and you add code. Add code and — with cheerful inevitability — you add the same code again, written slightly differently by someone who had no idea the first version existed. Left unsupervised, a codebase quietly becomes a slightly worse copy of itself. The bigger the team, the more of it is déjà vu.

And that was before the agents arrived. A coding agent writes more code, faster, with even less idea of what already exists, so a single engineer now ships the output of a whole team. The multiplier stopped being your headcount. It became how much code your headcount can summon.

Duplicated code → engineers + coding agents 1 dev 10 100 1,000+ devs
Illustrative, and yet suspiciously familiar. Duplication already grew faster than the team that produced it. Coding agents just removed the team as the limit.

Why DRY.codes

Save up to 80% in token usage

Every time an agent regenerates code that already exists, you pay for it twice — once to write it, and forever after to keep two copies politely agreeing with each other. DRY.codes gives your agent the memory to find and reuse what's already in your repos, so it writes less and ships more.

Regenerate from scratch 80% wasted re-writing existing code Find & reuse with DRY.codes ~20% −80% tokens saved
Illustrative: the red stripes are the tokens you burn re-writing code that already exists — exactly what DRY.codes lets your agent skip.

The knowledge base

Not a linter. A knowledge base for your code.

Indexing a repository doesn't just scan it, it adds the code to a connected knowledge base of how things are already built. Every file becomes a node; every shared pattern, a link. Your agent queries that graph over MCP to find what already exists across your repos (and the public corpus) before it writes a single new line. The more you index, the more the graph knows, and the less your agents repeat.

repo-a repo-b repo-c DRY .codes reuse duplicate
Illustrative: index your repos and they stop being islands. DRY.codes links the same function wherever it already lives, so your agent finds it (green) instead of writing the duplicate (amber) for the twelfth time.

How it works

From repo to reuse

  1. 1

    Connect your repos

    Sign in with GitHub and pick the repositories DRY.codes should index.

  2. Installrecommended

    The skill makes your agent check DRY.codes before writing — turning the AGENTS.md rule from the top of this page into something it follows every time.

    One command installs the skill and connects DRY.codes — sign in, then pick your repos or the public corpus:

    /plugin marketplace add max-network/plugins
    /plugin install dry-codes@max-network

    Register the marketplace, then install the dry-codes plugin (or enable it from Codex's plugin directory):

    codex plugin marketplace add max-network/plugins
    codex plugin add dry-codes@max-network

    Point any MCP client at https://dry.codes/mcp and sign in — then pick an endpoint or the public corpus. (A client without OAuth? Use a per-endpoint URL from the dashboard.) Then install the skill:

    curl -sL --create-dirs \
      -o .claude/skills/dry-codes/SKILL.md \
      https://dry.codes/skill/SKILL.md

    View the skill · works with Claude Code, Codex, Claude.ai, and the Agent Skills standard.

  3. 2

    Point your agent at it

    Sign in to https://dry.codes/mcp to Claude, Cursor, or any MCP client.

  4. 3

    Reuse, don't regenerate

    Your agent finds and reuses what already exists — fewer tokens, less duplication, a cleaner codebase.

Stop repeating yourself.

Index your first repository free. Your future self, and your token bill, will be quietly grateful.

Get Started — Sign in with GitHub View plans

Or point your MCP client at every public repo — .