Findings & checks · 1 flagged
MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · README.md
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
- Delete blanket "don't ask / no need to confirm" directives from the skill.
- If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.
Trace & refs
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
⚡ AutoCode
<div align="center">
You say it. AutoCode ships it.
🇨🇳 中文 · 🇯🇵 日本語 · 🇰🇷 한국어 · 🇪🇸 Español · 🇧🇷 Português · 🇩🇪 Deutsch · 🇫🇷 Français · 🇷🇺 Русский · 🇸🇦 العربية · 🇮🇳 हिन्दी
Every AI coding tool helps you write code. AutoCode helps you ship it.
>
You say what you want. AutoCode builds it, deploys it, and gives you a live URL. You don't need to know programming. You just say yes or no. When it's done, anyone in the world can visit your website.
>
New in v6: see a product you like, get your own. Say "I want something like X". AutoCode studies X from its public pages only, writes a blueprint where every line carries its source, cuts an MVP small enough to build in one session, and builds it. It learns the structure and designs its own look — nothing is reverse-engineered or copied.
>
Built on iLang, the first protocol to formally map Greek mathematical symbols (Σ, Δ, φ, λ, Ω, ∇, μ, Π, ψ, ξ, ζ, θ, ∂) as primitive verbs for AI-to-AI communication, and the first to define a computable vector space for AI judgment (11 dimensions, 4 axioms).
</div>
🚀 Install
Any agent, one command (Claude Code, Codex, Hermes, OpenCode, Gemini CLI, CodeBuddy, Qwen Code, Windsurf, Cursor, Trae, OpenClaw …):
git clone https://github.com/ilang-ai/autocode.git && cd autocode
python3 autocode.py setup # detects your agents, shows the plan, asks once, installs
python3 autocode.py doctor # read-only health check; run it whenever something feels off
python3 autocode.py uninstall # removes only what setup installed; your edits stayPython 3.8+ and nothing else. setup copies the 11 skills into each agent's skills folder, adds one managed block to its instructions file (CLAUDE.md, AGENTS.md, SOUL.md, …) so agents without hooks still get memory, full review and the pre-deploy secret scan, and keeps backups of anything it touched. python3 autocode.py clients lists the supported agents; the table lives in agents.json — adding an agent is adding an entry.
Claude Code plugin route (also fine):
/install autocodeThat's it. One command. View on Claude Code Marketplace
Or from GitHub:
/plugin marketplace add ilang-ai/autocode
/plugin install autocode@autocodeRestart. Start talking. Done.
Codex
Tell Codex:
Fetch and follow instructions from https://raw.githubusercontent.com/ilang-ai/autocode/refs/heads/main/.codex/INSTALL.mdOpenCode
Tell OpenCode:
Fetch and follow instructions from https://raw.githubusercontent.com/ilang-ai/autocode/refs/heads/main/.opencode/INSTALL.mdWorks on Claude Code, Codex, and OpenCode. Same 11 skills, zero changes.
Hermes Agent
hermes skills install ilang-ai/autocodeCursor
Copy skills/ into ~/.cursor/skills/, or browse Cursor Directory once listed.
Gemini CLI
gemini extensions install https://github.com/ilang-ai/autocodeAny SKILL.md-compatible agent
Copy the skills/ directory into your agent's skills path.
🔰 New to Everything? No Problem.
Never used a server before? Don't know what a terminal is? That's fine — this guide starts from zero.
<details> <summary><b>Click here for the complete beginner guide</b></summary>
Step 1: Get a Cloud Computer
You need a cloud computer (VPS) — think of it as a computer in the cloud that's always on.
👉 Go to [dev.ilang.ai](https://dev.ilang.ai) — new users get $300 free credit (enough for over a year).
How to sign up:
- Click the link → registration page
- Enter email, create password (10+ characters, mix of uppercase, lowercase, numbers)
- Check email, click verification link
- Payment: supports credit card, PayPal, Alipay (支付宝)
- $300 free credit applied automatically
Create your cloud computer:
- Click "Deploy +" (top right)
- "Deploy New Server"
- Type: "Cloud Compute — Shared CPU" (cheapest)
- Location: Closest to you (Tokyo for Asia, New York for Americas, London for Europe)
- Image: "Debian 12"
- Plan: $6/month (1 CPU, 1GB RAM — enough)
- Click "Deploy Now"
Wait 1-2 minutes. Green "Running" = ready.
Find your connection info:
Click server name → you'll see:
- IP Address — like
149.28.xxx.xxx - Username —
root - Password — click eye icon to show, copy button to copy
Step 2: Connect
Windows: Download PuTTY. Enter IP, click Open. Type username and password.
Mac: Open Terminal. Type: ssh root@your-ip-address. Enter password.
Phone: Download "Termius". Add host with IP, username, password.
Black screen with # = you're in.
Step 3: Install Everything
Paste this, press Enter, wait 2 minutes:
curl -fsSL https://raw.githubusercontent.com/ilang-ai/autocode/main/setup.sh | bashStep 4: Start Claude Code
claudeLog in when prompted (Anthropic account needed).
Step 5: Install AutoCode
Say:
Install the AutoCode plugin from github.com/ilang-ai/autocodePermission questions? Always choose the highest number ("Yes, don't ask again"). Safe — AutoCode is pure text, no executable code.
Type /exit, then claude to restart.
Step 6: Build
Build me a restaurant website with online reservationsYou're a developer now.
</details>
📖 What You'll Experience
First Time
You: "I want to sell cakes online"
AutoCode: "What computer do you use? Mac or other?"
You: "Other"
AutoCode: "I'll help you get a server. $6/month. OK?"
You: "OK"
AutoCode: → Sets up server → Builds website → Deploys
→ "Done. Open nainai-cake.com on your phone."
Your effort: 3 yes/no answers
Result: a live website anyone can visitFifth Time
You: "Build a booking site for my hair salon"
AutoCode: → Knows your preferences → Builds → Deploys
→ "Live at salon-booking.com. SSL active."Twentieth Time
You: "New project"
AutoCode: → Barely asks → Ships it → You realize you have 20 live websitesUser feedback: "I don't even know who did the work — I thought I was supposed to code, but it was already done."
🔍 See a product you like, get your own
You: "I want a lightweight issue tracker like Linear, for my 10-person team"
AutoCode: "I'll study Linear from its public pages, give you a blueprint of how
it's built, and cut an MVP we can ship today. We learn the structure
and design our own look."
→ reads pricing, docs, changelog, API docs, integrations, security, careers
→ blueprint.md: 36 features (core / plus / drop), 13 objects with relations,
7 user flows, pricing, gaps — every line tagged [S3] {high|medium|guess}
→ mvp.md: 5 objects, 7 screens, 7 build steps, 5 acceptance checks
You: cross out what you don't want, say "go"
AutoCode: → builds the MVP → deploys → "Live at https://… — open it on your phone."The product-study skill works from public pages only: official site, docs, pricing, changelog, public API docs, open-source repos, app-store listings, public reviews. It refuses login areas, captures, decompiling and bulk scraping on its own, and its gate rejects a blueprint that cites nothing, copies a sentence, or embeds the original's images, logo or code. The interface, copy, brand and code of your product are yours. A finished study: skills/product-study/examples/linear/. A complete run, study → build → verify → ship: skills/product-study/examples/dino/ — a non-technical ask ("a dinosaur game like that one, but I set the speed"), a one-file game whose rules are kept apart from its look, and an acceptance table the user can read (check / expected / actual / match, 8/8, rerun on the live URL).
Before shipping, AutoCode runs that table, not "tests passed". A static build ships to GitHub Pages at no cost; if you have no GitHub account it walks you through signing up, and you approve the login in your own browser. The whole journey, as the rules AutoCode follows, is methodology/see-it-ship-it.md.
🧠 11 Skills, 7 Stages
All automatic. You never activate them. Every skill is one stage of the journey or a support that runs across it; the journey itself is written down in methodology/see-it-ship-it.md.
| Stage | Skill | What it does |
|---|---|---|
| S1 see | see | Understands what you want from how you said it; your level; the size of the job; locks what you confirmed; asks at most one question that is yours |
| S2 study | product-study | "Like X" → public-source blueprint with evidence → MVP cut → verify runner |
| S3 plan | plan | Steps that each end in something you can click; minutes, not hours; core first; risks and costs flagged |
| S4 build | build | Simplest stack that works, one feature at a time, rules kept apart from the look, checkpoint before anything risky, first-session environment |
| S5 verify | verify | Silent checks after every feature, full review before save and deploy, the acceptance table — check / expected / actual / match — local and again on the live URL |
| S6 ship | ship | Secret scan, then a public URL: GitHub Pages at no cost for static builds, a server only when needed; domain if you want one |
| S7 hand-back | hand-back | Progress lines, real milestones credited to you, session summary, and the final hand-back: URL, table, what to try, what was learned vs decided |
| across | judge | When AutoCode acts alone and when it stops and asks — iLang v5.0: perception sets 11 dimensions, the frozen f_v5 sets the mode, you see one plain line |
| across | fix | Observe → reason → smallest fix → verify → explain in your words |
| across | memory | Loads what we know at every start, learns patterns, preferences and mistakes without asking, saves on stop — never secrets |
| across | security | Secrets never in code or chat; scan before every deploy |
v6.1 had 49 skills; 44 of them were one paragraph each. v6.2 folds them into the eleven above (the mapping is in the methodology file). setup retires the old folders on upgrade and keeps any you edited.
🎯 Precision
Powered by iLang v5.0. Structured instructions mean AI gets it right the first time. No rework.
🧠 Memory
Close today → open tomorrow → remembers everything. Gets better every session.
🌍 Different from Everything Else
| Other AI coding tools | AutoCode | |
|---|---|---|
| For | Developers | Anyone |
| Ends at | Code written | Website live |
| Requires | Terminal/IDE knowledge | Just say yes or no |
| Tech decisions | Asks you | Decides for you |
| Deployment | You figure it out | Automatic |
| Domain + SSL | You figure it out | Guided step by step |
| Start from a product you like | "Make a clone" and hope | Public-source blueprint with evidence → MVP |
| Memory | Resets each session | Persistent |
| Protocol | None | iLang v5.0 |
| Language | Usually English | Your language |
❓ Troubleshooting
<details> <summary><b>Something broke, don't know what to say</b></summary> Say "something's wrong". AutoCode asks simple questions. If it can't fix it in terminal, it tells you how to screenshot + upload to claude.ai. </details>
<details> <summary><b>/plugin command doesn't work</b></summary> Just say: "Install the AutoCode plugin from github.com/ilang-ai/autocode". Talking works better than commands. </details>
<details> <summary><b>Permission popups I don't understand</b></summary> Choose the highest number. AutoCode is pure text — safe to approve. </details>
<details> <summary><b>Can't connect to server</b></summary> Check IP, username, password in Vultr dashboard. Server must be green "Running". </details>
<details> <summary><b>Is it safe?</b></summary> 100% open source, pure text files. Claude Code verified: "No malicious content." </details>
<details> <summary><b>Cost?</b></summary> AutoCode: free forever (MIT). Cloud computer: $6/month (or free with $300 Vultr credit). Plus Anthropic API usage. </details>
<div align="center">
ilang.ai · ilang.cn · Hugging Face · @adsorgcn
iLang Research · iLang Inc. · Canada · MIT · Free Forever
You say it. AutoCode ships it. Powered by [iLang v5.0](https://ilang.ai).
</div>
Citation
CITATION.cff; Zenodo archives each release. Concept DOI 10.5281/zenodo.22865145 (all versions).