Fix One Vulkan C++ Test
SkillDev toolsBuild MLX, reproduce one failing C++ test, and implement the smallest Vulkan-only fix. Use when asked to repair a C++ test failure in this repository while preserving test code and avoiding unrelated fixes.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Fix One Vulkan C++ Test skill
What this skill tells your AI
The instructions your AI receives, as published by goniz/mlx-vulkan in .agents/skills/mlx-fix-test-cpp/SKILL.md and read by ahel’s review.
Fix exactly one C++ failure per invocation.
- Run
./dev.sh build. - After the build completes, run
./dev.sh test-cpp --fail-after=1. - Identify the first failing test and reproduce only that failure as needed.
- Do not modify test code. Modify implementation only under
mlx/mlx/backend/vulkan/. - Before editing, inspect corresponding CUDA and Metal implementations under
mlx/mlx/backend/cuda/andmlx/mlx/backend/metal/. - Make the smallest correct fix consistent with existing backend patterns and the async Vulkan pipeline policy.
- Run
./dev.sh build, then./dev.sh test-cpp --fail-after=1sequentially to verify the addressed failure. Continue only on the same failure until it passes or a clear blocker remains. - After implementation changes, run
./dev.sh generateto validate inference output coherence.
Report the failure addressed, Vulkan files changed, reference backend files consulted, and final test and generation results. Do not fix later or unrelated failures.
Signals
- GitHub stars
- 38
- Forks
- 5
- Last commit
- Aug 2026
Advanced
- Catalog kind
- skill
- Gateway key
mlx-fix-test-cpp- Source
- github.com/goniz/mlx-vulkan