The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the agent with specific shell commands using bash, find, and rg (ripgrep) to navigate, search, and manage the local Triton repository and execute setup scripts such as update-repos.sh and install.sh.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from an external Triton repository (TRITON_REPO), including tutorials, examples, and production kernels. This creates a vulnerability surface where malicious instructions hidden in source code or documentation could potentially influence the agent.
Ingestion points: Source files, tutorials, and examples located within the paths defined in TRITON_REPO (e.g., python/tutorials/, python/examples/).
Boundary markers: None identified; the instructions direct the agent to read and verify source surface content directly without explicit delimiters or warnings for embedded instructions.
Capability inventory: The agent is tasked with writing, debugging, and optimizing kernels, which implies file writing and potentially code execution capabilities.
Sanitization: No sanitization, escaping, or validation of the content read from the repository is specified.

triton-skill