Skip to content

Specialized Claude Agents

This project defines three specialized Claude agents in .claude/agents/. Each agent is pre-loaded with domain context to handle recurring tasks without needing to re-describe the project each time.

soap-tester

File: .claude/agents/soap-tester.md

Purpose: Generates unit and integration tests for SOAP operations.

Capabilities:

  • Given an operation name (e.g. send_movement_order), generates the corresponding XML response fixture in tests/fixtures/
  • Generates test_xml.py tests for build_* and parse_* functions
  • Generates test_operations.py tests with responses for HTTP mocking
  • Verifies that tests cover: happy path, returnValue != 0, SOAP fault, network error

adr-reviewer

File: .claude/agents/adr-reviewer.md

Purpose: Verifies consistency between ADRs and the implemented code.

Capabilities:

  • Reads ADRs and source code
  • Reports drift (e.g. external dependency added without an ADR, undocumented config parameter)
  • Proposes ADR updates if the code has evolved in a justified way

integration-checker

File: .claude/agents/integration-checker.md

Purpose: Analyses the impact of a change to the public interface.

Capabilities:

  • Given a proposed change to a HanelWarehouseGateway method signature, evaluates backward compatibility
  • Verifies that public dataclasses have no breaking changes
  • Verifies that introduced exceptions are subclasses of HanelGatewayError

Agent file structure

Each agent file follows this format:

---
name: <name>
description: <one-line description>
---

<system prompt with project context, constraints, expected output format>

Agents use only file read and write tools — they do not execute system commands. Update the agent files whenever the architecture changes.