The Agent Skills Directory

[DYNAMIC_EXECUTION]: The pattern file patterns/P6-decision-boundary.md contains the command np.load("data.npz", allow_pickle=True). Enabling allow_pickle is a known security vulnerability that allows for the execution of arbitrary code stored within a malicious .npz file during the loading process.
[COMMAND_EXECUTION]: The skill is designed to generate Python scripts based on user arguments and then execute them using the shell command uv run (Phase 5). This design creates a powerful capability for the agent to execute arbitrary code on the host system if the generation logic is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from $ARGUMENTS and external data files (CSV, NPZ) to generate and execute code, creating a surface for injection attacks.
Ingestion points: User-provided chart descriptions in $ARGUMENTS and external data files (e.g., data.csv, data.npz).
Boundary markers: None identified. The skill lacks explicit delimiters or instructions to treat user-provided data as non-executable text during the code generation phase.
Capability inventory: The skill has access to Write (to create scripts) and Bash (to execute them via uv run).
Sanitization: No sanitization or validation logic is present to filter malicious instructions from the user arguments or data files before they are interpolated into the generated Python code.

matplotlib