Design a programming language that coding agents cannot use, to test what they really understand
To check whether LLMs understand code or mostly recall it, a developer built KillSwitch, an esoteric language meant to be usable by humans but adversarial to agents. The spec hides lies a careful human would spot, rules depend on counting letters in identifiers, and phrases like "ignore previous instructions" have meaning in the language.
Source
dom96 on Hacker News · 2026-10-17 · Project link
Written in our own words. Credit to the original author. Link checked 2026-10-19.
What they did
To check whether LLMs understand code or mostly recall it, a developer built KillSwitch, an esoteric language meant to be usable by humans but adversarial to agents. The spec hides lies a careful human would spot, rules depend on counting letters in identifiers, and phrases like "ignore previous instructions" have meaning in the language.
Result (as claimed)
A public test bed for how coding agents cope with a language they have never seen; 221 points on Hacker News.
Tools they used
Cost: Claude Code: from $17/mo; OpenAI Codex: free plan, paid from $8/mo.
Set it up
Gather your sources
Collect the papers, notes, links or files you want to work from in one place.
Load them into the tool
Open Claude Code, OpenAI Codex and add your sources. Keep the original files so you can check claims later.
Ask focused questions
Ask for summaries, comparisons or connections with citations to the source passages.
Verify and save
Spot check claims against the sources, then save the output as notes you can reuse.
Copy the full setup prompt
Paste into your agent and it will walk you through the setup, asking before anything that sends, posts or spends.
You are helping me set up this AI use case end to end in Claude Code. Goal: Design a programming language that coding agents cannot use, to test what they really understand What the original author did: To check whether LLMs understand code or mostly recall it, a developer built KillSwitch, an esoteric language meant to be usable by humans but adversarial to agents. The spec hides lies a careful human would spot, rules depend on counting letters in identifiers, and phrases like "ignore previous instructions" have meaning in the language. Reported result (as claimed by the author): A public test bed for how coding agents cope with a language they have never seen; 221 points on Hacker News. Source: https://news.ycombinator.com/item?id=50135762 Tools they used: Claude Code, OpenAI Codex Category: Research & learning Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Codex. Goal: Design a programming language that coding agents cannot use, to test what they really understand What the original author did: To check whether LLMs understand code or mostly recall it, a developer built KillSwitch, an esoteric language meant to be usable by humans but adversarial to agents. The spec hides lies a careful human would spot, rules depend on counting letters in identifiers, and phrases like "ignore previous instructions" have meaning in the language. Reported result (as claimed by the author): A public test bed for how coding agents cope with a language they have never seen; 221 points on Hacker News. Source: https://news.ycombinator.com/item?id=50135762 Tools they used: Claude Code, OpenAI Codex Category: Research & learning Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Grok Bot. Goal: Design a programming language that coding agents cannot use, to test what they really understand What the original author did: To check whether LLMs understand code or mostly recall it, a developer built KillSwitch, an esoteric language meant to be usable by humans but adversarial to agents. The spec hides lies a careful human would spot, rules depend on counting letters in identifiers, and phrases like "ignore previous instructions" have meaning in the language. Reported result (as claimed by the author): A public test bed for how coding agents cope with a language they have never seen; 221 points on Hacker News. Source: https://news.ycombinator.com/item?id=50135762 Tools they used: Claude Code, OpenAI Codex Category: Research & learning Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
The Daily 5
Get five new use cases tomorrow morning
Free. Real sources. Unsubscribe any time.
Guides for this use case
How to vibe code an app with Claude Code · How to run multiple coding agents together · How to add memory to coding agents · How to build agent memory and context layers
Related use cases
Compare what AI coding agents really cost, with a pricing change log
A builder made a tracker that puts 18 coding tools (Cursor, Claude Code, Copilot, Codex and more) side by side and logs every pricing change with sources.
Give coding agents design inspiration from real websites
Built a design MCP server for Claude Code, Codex and OpenCode that searches hundreds of websites and feeds design inspiration into the agent.
Give coding agents a full Blender 3D production workflow with 20 skills
A repo of 20 Blender 5.2 skills walks an agent through modeling and sculpting, retopology and UVs, PBR materials and baking, geometry nodes, rigging, animation, rendering and export to Roblox, Unity, Unreal and Godot. A QA skill checks for z-fighting, floating parts, bad normals and UV issues before an asset counts as finished.
Run your coding agents as pixel workers in a browser office
The Agent Office shows Command Code, Claude Code, Codex and five more agents as pixel workers. You hire them into Daytona, Vercel or Railway sandboxes, drive them from the browser and approve actions with one key.