The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes the tilelang.compile() function to generate and compile GPU kernels from a Python DSL at runtime, facilitating the execution of dynamically generated code on hardware.- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the tile-kernels package via pip and references a public repository on GitHub hosted by DeepSeek.- [COMMAND_EXECUTION]: The documentation includes several shell commands for environment setup, package installation, and executing test suites using pytest.- [INDIRECT_PROMPT_INJECTION]: The library provides tools for Mixture of Experts (MoE) routing and activation quantization. If an agent bases its subsequent logic or instruction processing on the values produced by these kernels when they are operating on untrusted data, a theoretical injection surface exists.
Ingestion points: The tile_kernels/moe/ and tile_kernels/quant/ modules process input tensors that may originate from untrusted external sources.
Boundary markers: There are no explicit instructions or delimiters to isolate processed data from the agent's control flow.
Capability inventory: The skill possesses the ability to compile GPU kernels and execute shell commands for testing and benchmarking.
Sanitization: No input validation or tensor sanitization is described in the provided implementation examples.

tilekernels-gpu-kernels