Weekly Dev Log 2026-W20

This week marked a continuation of the ongoing development work on the Chord Tone Practice app. After a brief five-day break, the developer returned to Japan where autumn leaves the feeling of cooler weather, providing an ideal setting for picnics. The developer also shared the experience of planting lavender seeds with their children, eagerly observing their growth.

The primary focus during this week was the implementation of the backend logic for the Chord Tone Practice feature. This followed the completion of the UI design work in the previous update. By utilizing Codex, the developer efficiently executed the core models, calculation logic, and tests for Chord Tone Practice. Given that the requirements and implementation plan were already prepared, the initial implementation proceeded swiftly.

Subsequently, the developer thoroughly reviewed the backend logic using the ChordTonePracticeCalculatorTests created throughout the development process, cross-referencing with Swift documentation to ensure comprehension.

In addition to the development work, the developer continued their AI Security Learning Path on TryHackMe, specifically focusing on the AI System Reconnaissance room. This week's learning revolved around understanding how various exposed components, such as model registries, Jupyter notebooks, inference servers, and AI-related services, can collectively expose a broader attack surface than a single service might suggest.

The developer found this to be a more comprehensive week, involving not just feature implementation, but also code review and learning.

The key takeaway from a SwiftUI learning perspective was the importance of separating music-theory calculations from SwiftUI state, enabling easier testing and understanding of the core logic independently of the UI. The developer also appreciated how automated tests serve dual purposes: verifying code functionality and offering insights into understanding unfamiliar implementation logic.

The Swift concepts CaseIterable, Identifiable, and Hashable were reviewed in detail. Each of these played a distinct role in the development process - CaseIterable for the allCases enumeration, Identifiable for stable identity in SwiftUI components, and Hashable for utilizing values in collections and as dictionary keys. The developer also learned about the sorted(by:) closure, which defines the ordering of elements rather than moving them directly.

They learned about the usefulness of nonisolated for pure value models that do not depend on actor-isolated UI state.

On the security front, the developer observed that a single exposed AI-related service can reveal a significant amount of information about the system. They emphasized how AI systems can have a larger attack surface due to their complex interactions with specialized services, tools, models, and external dependencies. The developer also learned to connect individual reconnaissance findings into a larger picture of the system, and how model registries can expose metadata about an organization's models, environments, and dependencies.

They further studied AI supply chain risks, including exposed access tokens, external model sources, and poisoned dependencies. Lastly, they reviewed the MITRE ATLAS framework to better understand reconnaissance techniques against AI systems.