Exercise 4.1 -- Tasks and Knowledge

# Exercise 4.1 -- Tasks and Knowledge

**Context Engineering Layer: Persistence** -- Build task DAGs and extract tacit knowledge into files

## Goal

Design a persistence layer for the contract review agent so that knowledge survives across sessions and context compactions. You will create a task tracking system and extract tacit knowledge into persistent files that the agent can reference.

## Prerequisites

- Completed Module 3 (you have an architecture map and optimized CLAUDE.md)

## What You Have

- Your working agent from previous modules
- `knowledge-extraction.md` -- Template for capturing tacit knowledge (in this exercise folder)

## Key Concept: Persistence

The context window is volatile -- it empties when you `/clear` or start a new session. Anything the agent "learned" during a conversation is lost unless you persist it to files.

Persistence transforms:

- **Volatile knowledge** (exists only in conversation) into **durable knowledge** (written to disk)
- **Implicit understanding** (built up over turns) into **explicit instructions** (readable in a file)

## Your Tasks

### Step 1: Build a Task DAG for a Multi-Contract Review

Imagine you receive three contracts from the same client that need to be reviewed together. Create a `tasks.md` file that defines the task dependency graph: