The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/git4d_runtime_check.py executes diagnostic commands such as git, nvidia-smi, nvcc, and sysctl to gather environment information. These operations are implemented using safe subprocess list patterns, preventing command injection.
[PROMPT_INJECTION]: The skill processes external data from the Git history (commit subjects and author names) to create a graph visualization, which represents an indirect prompt injection surface.
Ingestion points: scripts/git4d_runtime_check.py extracts data from local git log output.
Boundary markers: Absent; commit messages are processed without explicit delimiters to distinguish them from instructions.
Capability inventory: The agent has access to bash, read_file, and write tools via the agents/openai.yaml configuration.
Sanitization: Commit data is structured into JSON before being provided to the agent, providing a layer of structural isolation.

git4d-runtime