PHIacademy

Autonomous Agentic Workflows — AI skill for Claude Code

AI community

MSc AI Capstone 6. Agentic AI Systems.

How to install Autonomous Agentic Workflows

This entry records only its repository, not the path inside it, so there is no exact command to give. Open PHIacademy/autonomous-agentic-workflows and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Autonomous Agentic Workflows does

MSc AI Capstone 6. Agentic AI Systems. Python, smolagents, LiteLLM, Claude Haiku 4.5, Pandas, Matplotlib. Covers multi-agent orchestration, tool-calling agents, conversation memory, adaptive HKDSE ICT question selection, least-privilege safeguards, and scenario-based evaluation with Jupyter Notebook. Includes APA-cited design report.

Alternatives in AI

Guizang Ppt Skill — AI-agent Skill for generating polished HTML slide decks: editorial magazine and Swiss layouts, image prompts 24.9k ★
Cc Mirror — Create multiple isolated Claude Code variants with custom providers (Z.ai, MiniMax, OpenRouter, LiteLLM) 2.1k ★
OpenMobius Skill — ICT/SMC trading-knowledge skill for AI coding agents (Claude Code / Codex / OpenClaw / Hermes) 660 ★

README

MSc in AI Capstone #6: Agentic AI Systems – Adaptive HKDSE ICT Practice Tutor

Repository: https://github.com/PHIacademy/autonomous-agentic-workflows.git

A multi-agent adaptive practice tutor, built with smolagents and Claude Haiku 4.5, that serves, grades and explains past-paper multiple-choice questions from the Hong Kong Diploma of Secondary Education (HKDSE) ICT Paper 1A examination. This project implements an orchestrator with four specialised worker agents and 16 deterministic tools, and a scenario-based, evidence-based evaluation of how well its routing, grading and safeguards hold up.

Overview

Task: Design, implement and evaluate an agentic AI workflow (orchestrator plus specialised workers, with tools, memory and safeguards)
Domain: HKDSE ICT Paper 1A past-paper practice (adaptive question selection, grading, explanation, difficulty analysis)
Dataset: 560 real HKDSE ICT Paper 1A multiple-choice questions (2012–2025), published by the Hong Kong Examinations and Assessment Authority (HKEAA), distributed via a Kaggle dataset prepared for this capstone. (same dataset as Capstone #5)
Framework: smolagents ToolCallingAgent with managed agents
Model: Claude Haiku 4.5, accessed through LiteLLM
Architecture: 1 Orchestrator + 4 workers (Question Selector, Progress Tracker, Explainer, Question Analyst) and 16 tools
Headline design choice: safeguards are enforced by tool scope (e.g., the answer key is reachable only by the grading tool), not by prompt instructions alone

Project Structure

.
├── agentic_system.ipynb        # Main notebook: agents, tools, example runs, analysis
├── architecture_diagram.png    # System architecture (orchestrator, workers, tools, data)
├── requirements.txt            # Python dependencies
├── env.example                 # Template for API credentials (copy to .env)
└── data/
    ├── final_dataset.csv       # Question bank (on