workshop-harness AGENTS.md

AGENTS.md - Agent Instructions for Workshop Harness

Standard AI Agent Context & Instructions Specification (https://agents.md/)

This repository provides seamless integration for all AI Coding Agents including Google Antigravity, Gemini CLI, Anthropic Claude Code, OpenAI Codex / ChatGPT, Aider, and Cursor / Windsurf.

---

🎯 Repository Overview & Purpose

workshop-harness is an AI Agent Harness, Skill Suite, and CLI Automation Toolkit for technical workshops (Build with AI, DevFest, community coding labs).

• CLI Automation: python3 harness_cli.py
• 20 Agent Skills: Located under skills/*/SKILL.md
• Document Templates: Located under templates/doc-templates/

---

🤖 Natural Language Prompt ➔ Autonomous Skill Execution Rules

CRITICAL RULES FOR ALL AI CODING AGENTS (Antigravity, Claude Code, Codex, Cursor, Aider):

1. 🌐 Mandatory Autonomous Pre-Flight Web Search (workshop-web-researcher):
   • NEVER rely solely on LLM internal parametric training data when generating workshop curriculum, code samples, Colab notebooks, or setup guides. Parametric weights contain outdated/deprecated APIs and retired model tags.
   • You MUST AUTONOMOUSLY execute live web searches (search_web / web search) BEFORE writing any code or documentation to verify the newest SDK syntax (e.g. google-genai SDK, com.google.genai:google-genai-kotlin-android, google_generative_ai, genui), current model tags (e.g. gemini-3.7-flash, claude-sonnet-5, gpt-5.6-sol, gemma4), and breaking changes.
2. 🌍 Developer-Idiomatic Localization Protocol (No Literal Translations):
   • When translating or authoring slide decks, setup guides, or runbooks into Korean (KR), Japanese (JA), or Chinese (ZH), NEVER perform awkward literal / word-for-word machine translations.
   • MUST use native developer jargon and community-standard technical terms used by real-world software engineers:
     • Korean (KO): Use natural IT developer terms (e.g., "프롬프트 엔지니어링", "핸즈온 랩", "스캐폴딩", "의존성 주입", "양자화", "파인튜닝", "런북", "임베딩"). Never use awkward literal phrasing (e.g. ❌ "손으로 하는 실험실", ❌ "달리는 책").
     • Japanese (JA): Use standard developer Katakana / technical terms (e.g., 「ハンズオン」, 「デプロイ」, 「スキャフォールディング」, 「トークン」, 「ファインチューニング」, 「ランブック」).
     • Chinese (ZH): Use Chinese tech community terminology (e.g., 「实战工作坊/动手实验」, 「脚手架」, 「提示词工程」, 「微调」, 「演练手册/Runbook」).
3. 🚫 Typography Rule (No Em Dash):
   • NEVER generate em dash characters (— or \u2014) in any generated documentation, code comments, slide decks, or markdown files.
   • Always use standard hyphens (-), colons (:), or parentheses () instead.
4. 🚫 No Emojis in Documentation (Default: Clean Professional Text):
   • Do NOT include emojis in generated documentation, slide decks, setup guides, code comments, or runbooks by default.
   • Only use emojis if the user explicitly requests them in their prompt (e.g., "use emojis in headings").
5. 🗣️ Natural Language Autonomy:
   • Users communicate via natural language chat prompts. Do NOT instruct users to run Python commands themselves.
   • When a user requests a task, you MUST autonomously read the relevant skills/*/SKILL.md and execute the appropriate harness CLI command or generate the code artifacts on behalf of the user.