en:authors_main:task_editor:ai_agent

Skills for an AI agent: working with the whole game

The prompt page is for a chatbot that writes tasks in a chat. This one is for an agent with access to files (Claude Code, Codex, Cursor, Gemini CLI, Grok Build and the like): it takes the game downloaded from the editor, understands it and edits it — levels, codes, bonuses, hints, the order of tasks — not by reading the JSON itself but through a tool. It can also design a level from the engine's reference, with examples from the demo game.

Two skills:

  • qeng-game-json — reading and editing the game file: “what is on level 5”, “change the autopass”, “move level 7 to position 3”, “what is wrong with the game”.
  • qeng-game-author — how a level is written: the engine's reference, examples from the demo game, step-by-step recipes.

Give the agent the link to this page and ask it to install the skills — it does the rest by the instruction at the end of the page: downloads the archive, puts the skills where they belong, checks the version. It needs only Python 3.9 or newer; if the agent says there is none, install it from python.org.

Then: “Download JSON” in the editor — that file, game_N.json, is what you hand to the agent. Back it goes through the editor's import: “Save…” lists every change and the errors before writing.

The archive with both skills: qeng-game-skills.zip. It holds two folders, qeng-game-json and qeng-game-author; put both where your agent looks for skills.

The folders go into .claude/skills/ of the project (for that project only) or into ~/.claude/skills/ (for every project). A new skill is picked up on the next start; call it explicitly with /qeng-game-json.

All four read the same place: .agents/skills/ of the project, or ~/.agents/skills/ in your home folder.

  • Codex (CLI, IDE extension, app): call it explicitly with $qeng-game-json.
  • Cursor — from version 2.4; call it explicitly with /qeng-game-json in the agent chat.
  • Gemini CLI asks for permission before it turns a skill on; the list is gemini skills list.
  • Grok Build (xAI's terminal agent): besides .agents/skills/ it reads ~/.grok/skills/.

In claude.ai a skill is uploaded in the settings, and each one as its own archive: qeng-game-json.zip and qeng-game-author.zip.

  1. Turn code execution on: Settings → Capabilities. On Team and Enterprise the organization owner turns it and skills on.
  2. Customize → Skills → “+” → Create skill → Upload a skill — both archives, one after the other.
  3. Attach the game file to a message in the chat; Claude hands the edited one back as a file to download.

An agent without skill support manages too, as long as it can use files and a command line: unpack the archive next to the game and ask the agent to read qeng-game-json/SKILL.md and qeng-game-author/SKILL.md before working. Chatbots that cannot run code (Grok's web chat, for example) cannot use the skills: the tool needs Python.

You are installing two skills for working with qeng.org games (version 0.1.0):
- qeng-game-json: read and edit a game export - the JSON the game editor downloads - as objects (levels, codes, bonuses, hints, lines), through a Python tool, never by opening the JSON;
- qeng-game-author: how a level is written - the engine's reference by topic, working levels of the demo game as examples, step-by-step recipes.

1. Requirement: Python 3.9 or newer (`python --version` or `python3 --version`). Nothing else: the tool uses only the standard library.
2. Download https://wiki.qeng.org/lib/exe/fetch.php?media=authors_main:task_editor:qeng-game-skills.zip and unpack it. It holds two folders, qeng-game-json/ and qeng-game-author/. Put both where your agent looks for skills:
   - Claude Code: .claude/skills/ of the project, or ~/.claude/skills/ for every project.
   - Codex, Cursor (2.4 or newer), Gemini CLI, Grok Build: .agents/skills/ of the project, or ~/.agents/skills/ for every project - all four read it.
   - Any other agent: anywhere; read qeng-game-json/SKILL.md and qeng-game-author/SKILL.md before working.
   Some agents load new skills only on the next start, or ask the user to approve a skill first (Gemini CLI).
3. Verify: `python qeng-game-json/qeng_game.py --version` (run from the skills folder) prints `qeng_game.py 0.1.0`. If this page names a newer version than the one installed, download again.
4. The game: the author downloads it in the game editor ("Download JSON", a file game_N.json) and gives it to you. Work on that file only with qeng_game.py - start with `python qeng-game-json/qeng_game.py --game game_N.json summary`, then `levels`. When done, the author imports the file back in the editor; "Save..." there lists every change and checks it before writing.
5. Translating a game is not part of these skills: the editor's "Translate" tab does that.

This page as plain text: https://wiki.qeng.org/doku.php?id=authors_main:task_editor:ai_agent&do=export_raw

<note tip>Translating a game is not part of the skills — the editor's “Translate” tab does that; see Game localization.</note>

  • en/authors_main/task_editor/ai_agent.txt
  • Last modified: 2026/10/10 08:01
  • (external edit)