v2026.08.26 - 17 Skills Suite, Google ADK Multi-Agent Builder, ELI5 Explainer, Gemini 3.7 & Live

What's Changed in v2026.08.26

🚀 Key Highlights & New Capabilities

• 17th Agent Skill: eli5-concept-explainer:

  • Translates complex AI mechanics, model architectures, and runtime error signatures into 3-tier ELI5 (Explain Like I'm 5) physical analogies, visual Mermaid mental maps, and beginner cheat sheets.
  • Equips facilitators and TAs with 10-second live Q&A crisis scripts to eliminate cognitive overload for novice attendees.

• 16th Agent Skill: adk-workshop-builder:

  • Builds multi-language (Python, TypeScript, Go, Kotlin) autonomous agent & multi-agent system workshops using Google Agent Development Kit (ADK).
  • Standardizes Coordinator-Worker delegation, custom tool bindings, Human-in-the-Loop (HITL) checkpoints, and trajectory evaluation.
  • Generates language-specific starter vs final solutions (package.json, go.mod, build.gradle.kts, pyproject.toml).

• Gemini 3.7 Flash & Google GenAI SDK Support:

  • Integrated gemini-3.7-flash as the primary flagship workhorse model for coding and agentic tasks.
  • Added support for dynamic thinking budget control (thinking_config / thinkingLevel).
  • Standardized on unified google-genai Python and @google/genai JavaScript/TypeScript SDKs.

• Gemini Live API & Real-Time Multimodal Streaming:

  • Added full support for gemini-3.1-flash-live-preview (low-latency voice/audio/video) and gemini-3.5-live-translate-preview.
  • Added pub.dev gemini_live Flutter package integration (zero Firebase dependency, direct WebSocket streaming).
  • Added 3-step Gemini Live real-time curriculum track to hands-on-curriculum-builder.

• Next-Gen Model Specifications:

  • Added Anthropic Claude Generation 5 (claude-sonnet-5, claude-opus-5, claude-fable-5).
  • Added OpenAI GPT-5.6 series (gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna).

• 17 Agent Skills Metadata & Localization Sync:

  • Synchronized 17 skills metadata across docs/, README.md, CLAUDE.md, AGENTS.md, and website/app.js (EN, KO, JA, ZH).

v2026.08.16 — Google Colab CLI, Marp Slide Generator & 15 Agent Skills

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

---

[v2026.08.16] - 2026-08-16

Added

• 14th Agent Skill: colab-workshop-integrator:
  • Automatically transforms workshop labs and starter/final code into interactive Google Colab Jupyter Notebooks (.ipynb).
  • Injects zero-setup cells: automated %pip install, GPU accelerator verification (!nvidia-smi), and Colab Secrets credential handling (google.colab.userdata).
  • Generates 'Open in Colab' SVG badges and Colab README documentation.
  • Integrates with official Google Colab CLI (colab-cli) for automated headless cloud execution and smoke testing.
• 15th Agent Skill: workshop-slide-generator:
  • Generates publication-ready presentation slide decks in Marp Markdown (output/slides/slides.md) and standalone zero-dependency interactive Web Presentations (output/slides/index.html).
  • Syncs 1:1 with facilitator timeline markers in RUNBOOK.md.
  • Supports instant PDF/PPTX compilation via Marp CLI.
• CLI Commands (harness_cli.py):
  • Added build-slides command with --target, --output, and --export-pdf flags.
  • Added export-colab command with --target, --output, --repo, and --test flags.
  • Added test-colab command for one-click Colab CLI verification.
  • Updated generate-all pipeline to full 8-step orchestrator across all 15 skills.
• Documentation & Agent Standards:
  • Expanded AGENTS.md, multi-language READMEs (EN, KR, JA, ZH), and MkDocs documentation to 15 specialized skills.

---

[v2026.08.01] - 2026-08-01

Changed

• Full English Localization:
  • Fully translated all 12 SKILL.md instruction files to English to ensure global AI agent compatibility across Claude Code, Aider, Cursor, Codex, and Antigravity.
  • Translated all core document templates in templates/doc-templates/ (Hardware & Env, Prerequisites, Session Guide, Prompt Pack, Troubleshooting FAQ, Facilitator Runbook, FAQ Generator, Persona Loop Review) to English.
  • Updated python scripts (generate_prep_pdf.py, verify_workshop.py) and harness_cli.py docstrings and generated template strings to English.

---

[v2026.07.30] - 2026-07-30

Added

• 12 Specialized Agent Skills for end-to-end workshop automation:

  • workshop-scaffolder: Standard repository structure scaffolding
  • cross-architecture-checker: Apple Silicon, Intel Mac, Windows, Linux compatibility audit & fallback matrix
  • prerequisite-checker: OS-specific SDK/language installation guides, Google AI Studio & GCP Vertex AI API Key issuance steps, and automated check_env.sh/ps1 verification scripts
  • hands-on-curriculum-builder: Step-by-step lab curriculum, starter vs final code, prompt packs with mandatory references protocol
  • pdf-handout-generator: ReportLab + PyMuPDF based PDF handout builder with contact sheet preview
  • workshop-troubleshooter: RAM-tier (8G/16G/32G+) and OS-specific troubleshooting matrix
  • workshop-runbook-generator: Minute-by-minute facilitator & TA timeline runbook (RUNBOOK.md)
  • live-debug-assistant: 10-second terminal error hotfix diagnosis & API Key security protocol
  • workshop-faq-generator: Attendee FAQ auto-generation for hardware, network, and code questions
  • workshop-tester: Code execution smoke tests & markdown broken relative link auditor
  • workshop-web-researcher: Live web search for latest tool/SDK/AI model release tags & breaking change prevention
  • workshop-persona-loop-evaluator: Loop Engineering multi-persona review for beginner, intermediate, and advanced attendees

• CLI Automation Tool (harness_cli.py):

  • generate-all: One-Click Full Orchestrator triggering all 12 skills in sequence
  • init: Workshop project scaffolding with templates, scripts, and directory structure
  • audit-compat: Cross-architecture tech stack risk auditing
  • audit-loop: Loop Engineering multi-persona evaluation
  • test: Code smoke tests & markdown link integrity verification
  • build-pdf: Publication-ready PDF handout generation

• Astral uv Integration:

  • pyproject.toml for declarative dependency management
  • Auto-install of reportlab, pymupdf, pillow via uv on first run
  • All CLI commands support uv run harness_cli.py execution

• Open AI Agent Standard:

  • AGENTS.md adopting the open agents.md specification for vendor-agnostic agent interoperability
  • CLAUDE.md for Anthropic Claude Code CLI users

• Dynamic AI Model Discovery Protocol:

  • Prevents legacy model hardcoding (gpt-3.5, llama-2, gemma-1) via mandatory web search verification
  • Timestamped currency labels on all generated documentation

• Google AI Studio & GCP Vertex AI Key Issuance Guide:

  • Step-by-step Gemini API Key issuance via AI Studio
  • GCP Service Account JSON key generation for Vertex AI
  • .env.sample and .gitignore security enforcement

• Mandatory References Protocol:

  • All generated documents include References section with official documentation URLs and source citations

• Multilingual README Support:

  • README.md (English, primary)
  • README_KR.md (Korean)
  • README_JA.md (Japanese)
  • README_ZH.md (Chinese)

• Repository Banner Image (assets/workshop_harness_banner.jpg)

• Document & Script Templates (8 doc templates, 10+ script templates)

• Example Project (examples/one-click-demo-workshop/)

References

• Build with AI Seoul 2026
• Build with AI Golang Korea 2026
• Build with AI Mongo 2026

v2026.08.01 - Full English Localization

Workshop Harness v2026.08.01

Highlights

• Full English Localization:
  • Translated all 12 SKILL.md agent skills to English for seamless interoperability across international AI coding agents (Claude Code, Cursor, Aider, Antigravity, etc.).
  • Translated core workshop document templates in templates/doc-templates/ to English.
  • Standardized CLI scripts and docstrings to English.

Full Changelog

See CHANGELOG.md

v2026.07.30 - Initial Release

Workshop Harness v2026.07.30

Highlights

• 12 Specialized Agent Skills for end-to-end workshop automation
• One-Click Full Orchestrator (generate-all) triggering all 12 skills in sequence
• Astral uv Integration with pyproject.toml and auto-dependency installation
• Open AI Agent Standard via AGENTS.md (agents.md) & CLAUDE.md
• Dynamic AI Model Discovery Protocol preventing legacy model hardcoding
• Google AI Studio & GCP Vertex AI Key Issuance Guide in prerequisites
• Loop Engineering Multi-Persona Evaluator for beginner, intermediate, and advanced attendees
• Mandatory References Protocol for all generated documentation
• Multilingual README (English, Korean, Japanese, Chinese)

12 Agent Skills

#	Skill	Role
1	workshop-scaffolder	Standard repo structure scaffolding
2	cross-architecture-checker	Apple Silicon, Intel Mac, Win, Linux compatibility audit
3	prerequisite-checker	OS-specific SDK/API Key installation guide
4	hands-on-curriculum-builder	Step-by-step lab curriculum & prompt packs
5	pdf-handout-generator	PDF handout builder via ReportLab
6	workshop-troubleshooter	RAM-tier & OS troubleshooting matrix
7	workshop-runbook-generator	Facilitator & TA timeline runbook
8	live-debug-assistant	10-second terminal error hotfix
9	workshop-faq-generator	Attendee FAQ auto-generation
10	workshop-tester	Code smoke tests & broken link auditor
11	workshop-web-researcher	Latest tool/SDK/AI model version verification
12	workshop-persona-loop-evaluator	Loop Engineering multi-persona review

Quick Start

git clone https://github.com/JAICHANGPARK/workshop-harness.git
cd workshop-harness
./scripts/install_skills.sh
uv run harness_cli.py generate-all --name "my-workshop" --topic "Local RAG with Gemma 4" --stack "python,ollama,docker"

Full Changelog

See CHANGELOG.md

References

• Build with AI Seoul 2026
• Build with AI Golang Korea 2026
• Build with AI Mongo 2026