# Minecraft Mod QA Automation > Minecraft Mod QA Automation - Skill: `tim-osterhus/minecraft-mod-qa-automation` (Agent Skill) - Install (CLI): `npx skillmds@latest add tim-osterhus/minecraft-mod-qa-automation` - Raw SKILL.md: https://api.skillmd.com/api/skills/tim-osterhus/minecraft-mod-qa-automation/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tim-osterhus (https://skillmd.com/u/tim-osterhus) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tim-osterhus/minecraft-mod-qa-automation --- # Minecraft Mod QA Automation ## Purpose Help agents design strong automated QA for Minecraft Java mods, especially when compile-time checks and server-only smoke tests are not enough. Treat Minecraft mod QA as a layered testing problem: white-box runtime checks first, structured client bridge assertions second, and black-box desktop or image automation last. ## Quick Start 1. Run cheap checks first: formatting, build, unit tests, and loader-native tests. 2. Use Fabric GameTests or equivalent runtime tests for deterministic world interaction coverage. 3. Add a client bridge for command execution, player state, nearby block scans, camera control, logs, and screenshots. 4. Use Mineflayer or pathfinding only for bounded gameplay scenarios where movement, crafting, or repeated interaction adds signal. 5. Use native RPA or image matching only for GUI-visible behavior that cannot be asserted through commands, GameTests, bridge calls, or protocol state. ## Operating Constraints - Do not treat every mod check as UI automation. - Do not rely on Mineflayer alone and call it real client QA. - Do not launch the client for tiny logic checks that unit tests can cover. - Do not compare screenshots without fixed client, resource pack, GUI scale, camera pose, FOV, window geometry, and display scaling. - Do not use public multiplayer servers for reproducible QA. - Do not let automated play improvise without named setup, actions, checkpoints, assertions, and cleanup. - Do not silently retry real logic failures. Retry only plausible flakes such as lost focus, unloaded chunks, startup races, or screenshot timing. - Keep all automation constrained to local test worlds, private servers, or dedicated QA environments. ## Inputs This Skill Expects - Minecraft version, mod loader, Java version, and Gradle wrapper command. - The supported mod matrix and whether the target is Fabric, Forge, NeoForge, or another loader. - The feature under test and its expected runtime, gameplay, and visual states. - Available test hooks: unit tests, GameTests, commands, bridge mod, screenshots, prepared worlds, Mineflayer scripts, or desktop automation. - Host constraints such as WSL/Windows split, client launch method, ports, window geometry, resource packs, and artifact directories. ## Output Contract - Name the lowest reliable test layer that can prove the behavior. - Define a deterministic fixture: seed, save, setup commands, configs, mod set, resource pack state, camera pose, and cleanup/reset path. - Specify the exact observe-and-assert method: unit assertion, GameTest, command, bridge state query, screenshot comparison, bot scenario, or RPA step. - Require an artifact bundle with logs, screenshots where relevant, state dumps, action traces, environment data, and a structured test summary. - Classify failures as setup, launch/runtime, deterministic gameplay, visual, timing/flaky, unsupported test design, or unknown. ### Recommended Architecture For WSL plus Windows setups, keep orchestration and repository work on the WSL side, and keep the actual Minecraft client, native desktop automation, and client-side helpers on the Windows side. Recommended control flow: ```text WSL agent -> local launcher/helper -> Minecraft client -> bridge/test mods -> artifacts -> WSL judge ``` Do not force visual validation through headless Linux when the real target is a Windows client workflow. Use headless launchers for smoke coverage where they fit, not as proof that the real rendered client works. ### Layer Selection - Unit tests: logic, math, serialization, recipe eligibility, inventory transforms, and utility behavior. - Fabric GameTests or loader-native tests: world interaction, block placement, machine progression, item behavior, persistence, and deterministic runtime assertions. - Client bridge: command execution, player info, inventory and container state, nearby block scans, chat or command output capture, camera positioning, and screenshots. - Mineflayer: bounded protocol-level gameplay scenarios involving movement, crafting, placement, gathering, or repeated interaction. - Pathfinding or Baritone-like tooling: traversal, reachability, repeated navigation, or mining/building routes. - Screenshot helpers: chunk-settled capture timing, naming discipline, visual baselines, and bulk evidence. - PyAutoGUI, SikuliX, Java Robot, or similar RPA: exact GUI clicks, modal dialogs, HUD alignment, menus, and other behavior the game does not expose through a better hook. ### Test Classes - Startup smoke: clean launch, mod load, no fatal exceptions, environment ready. - Registration smoke: blocks, items, entities, screens, recipes, and commands exist and are usable. - Interaction tests: place, use, open, insert, consume, produce, break, reload, and verify state or persistence. - Visual assertion tests: HUD, overlay, tooltip, model variant, block state, particle, progress bar, or screen layout appears as expected. - Scenario tests: multi-step gameplay flow with named fixture, action trace, checkpoints, screenshots, and terminal assertion. - Regression replay: rerun a previously failing setup with the same seed, save, commands, scripts, and expected artifacts. - Compatibility tests: only the loader, Minecraft, Java, and dependency combinations the mod actually supports. ### Artifact Contract Every automated QA run should emit a structured artifact bundle. Minimum contents: - `run_manifest.json` - `environment.json` - `client.log` or extracted run log - `test_summary.json` - `screenshots/` when client-visible state matters - `failures/` - `commands_or_actions.log` - `world_state/` dump or fixture reference - `diffs/` when visual comparison is used Useful manifest fields include mod version, git SHA, Minecraft version, loader version, Java version, active mods, configs, world seed or fixture ID, scenario name, start/end timestamps, pass/fail, failing step, screenshots, and exception summary. ## Procedure 1. Define the feature behavior and pick the lowest reliable test layer. 2. Normalize the environment: Minecraft, loader, Java, mod set, configs, resource packs, render distance, GUI scale, FOV, keybinds, world seed, window size, display scaling, ports, and screenshot directory. 3. Run cheap checks first and stop if they fail. 4. Launch the client only after the basic checks pass. 5. Verify bridge/test helpers loaded, then enter a prepared world or fixture. 6. For each scoped test, run setup, act, observe, assert, capture artifacts, and cleanup/reset. 7. Escalate only when needed: GameTest or command assertion, bridge state query, screenshot comparison, Mineflayer scenario, then native RPA. 8. On failure, capture screenshot, player state, nearby block scan if relevant, inventory/container state, recent log tail, and failure class. ## Pitfalls And Gotchas - A cinematic bot demo is not stronger than a boring deterministic harness. - Screenshots without environment normalization are weak evidence. - Protocol-level bot state does not prove the true rendered client experience. - Client bridge inspection is usually more reliable than pixel clicking. - Long 20-minute scenarios should be split into smaller named tests. - Fixed seeds, prefab structures, prepared saves, or explicit setup commands are more useful than improvised worlds. - Visual tolerances need a reason, and full original screenshots should still be kept as evidence. ## Progressive Disclosure Start with the cheapest reliable layer and widen only when the feature needs runtime, client, visual, or gameplay proof. Keep the QA plan as small as the feature allows, but require enough artifacts that a future agent can reproduce a failure from a clean checkout or documented fixture. ## Verification Pattern - Confirm the plan names the test layer and explains why lower layers are not enough. - Confirm the fixture and environment are deterministic. - Confirm each test has setup, action, observation, assertion, artifacts, and cleanup/reset. - Confirm screenshots are paired with normalized settings and structured state where possible. - Confirm failures are classified instead of all being labeled mod bugs. - Confirm RPA is justified as a last resort, not the default testing strategy.