Task Automation AI Agent — AI skill for Claude Code

Task Automation AI Agent

A Python agent that breaks a natural-language request into sub-tasks and calls tools (calculator, knowledge search, file lookup) via **function calling** to complete multi-step tasks automatically — matching the resume line:

"Built a Python-based agent that breaks a user's natural-language request into sub-tasks and calls tools via function calling to complete multi-step tasks automatically. Designed the agent's decision loop to select the right tool for each sub-task and combine results into a single coherent response."

How it works

User request
     │
     ▼
 [ LLM / planner ]  ──decides──▶  which tool(s) to call + arguments
     │                                   │
     │                                   ▼
     │                          [ Tool execution ]
     │                          calculator / knowledge_search / file_lookup
     │                                   │
     ◀───────────── tool result ─────────┘
     │
     ▼
 Loop until no more tools are needed → final answer

Files

File	Purpose
tools.py	The 3 tools the agent can call, plus their JSON-schema definitions for function calling
agent.py	Real agent — uses Claude's tool-use API to plan and call tools. Requires ANTHROPIC_API_KEY.
demo_offline.py	Offline demo — same plan → call tool → respond loop, using simple keyword rules instead of an LLM. Runs with no API key, no internet.
data/	Sample files used by the file_lookup tool

Running it

**Offline demo (no API key needed) — good for a quick interview demo:**

pip install -r requirements.txt
python3 demo_offline.py

Try: `What is (12 + 8) * 3?`, `Tell me about agentic ai`, `Read file notes.txt`

**Real LLM-driven agent (requires an Anthropic API key):**

export ANTHROPIC_API_KEY=your_key_here
python3 agent.py

This version actually lets Claude decide which tool(s) to call and in what order, based on the request — t