mcdk-dev

Start with read-only observation: capture the game window and read the latest logs to gather evidence. Then run minimal execute_code probes to verify state, and only reload the game or addon after preserving screenshots and logs of the current scene.

Use capture_game_window to take a screenshot, estimate the target's percentage coordinates from a stable anchor point, then call click_game_window. Take another screenshot immediately after each click since coordinates become invalid when the UI changes pages.

Use reload_game when Python hot-reload is insufficient or logic state needs refreshing. Use reload_addon_and_game when resource packs, behavior packs, or addon configuration changed and need a full refresh, since it is heavier and disrupts the current scene.

get_latest_error_logs can omit certain errors such as some JSON parsing failures, so it lacks full context. When troubleshooting, default to get_latest_logs first and use get_log_range to examine the surrounding context of suspicious entries.

Switch to hypothesis-driven instrumentation when one screenshot and one log cannot explain the issue, when the problem involves timing, races, or intermittent behavior, or when you need persistent structured NDJSON logs and explicit cleanup of debug traces.