Building Specialized Agents

In our previous lesson, we discussed the theory of agent orchestration. We learned that using one large AI for a long project often leads to context decay, where the AI starts making mistakes because it is trying to remember too much at once. The solution is to use a Main Agent that delegates work to specialized Subagents. In this lesson, we move from theory to tools. We will learn how to actually build these specialized agents. On CodeSignal, these agents are defined as TOML configuration files stored in a specific folder: .codex/agents/. Each role is registered in your .codex/config.toml, then its detailed settings live in a separate .toml file. By creating these files, you are giving Codex a set of specialized "employees" it can hire to perform specific tasks with high precision.

In Unit 1 you already registered an agent role in .codex/config.toml and configured it with a role-specific TOML file. Every specialized agent follows that same three-part structure: Identity (the registration entry and model settings), Role (the developer_instructions system prompt that defines what the agent does), and Completion Criteria (a Standards section inside those instructions that lists what must be true before the agent reports "done"). We will apply this pattern directly to two new specialists below.

The Task Executor is your primary worker. Its job is to take a single task from your tasks.md file and turn it into working, tested code. You already saw its basic structure in Unit 1. What makes it effective in practice is the Process section of developer_instructions, which enforces a Test-First workflow—writing the test before the code to ensure we build exactly what is needed:

## Process
1. **Understand Task**
   - Read acceptance criteria from the spec.
2. **Test-First Implementation**
   - Write failing tests.
   - Verify they fail.
   - Implement code to make them pass.
3. **Self-Validate**
   - Run tests and check for type errors.

After the process, we define the Standards. These are the "laws" the agent must follow, such as requiring 90% code coverage. This ensures that even if you aren't watching, the agent maintains high quality.

## Standards
- Coverage ≥90%
- All tests pass
- No type errors (using mypy)

This agent can be summoned by asking Codex to: Spawn a task-executor agent: "Execute T001: Add priority field to Task model".