The Agent Skills Directory

[Prompt Injection] (SAFE): No evidence of instructions attempting to bypass safety filters or override agent behavior. The markdown files provide educational theory and usage examples without malicious intent.
[Data Exposure & Exfiltration] (SAFE): Scripts do not import network libraries (requests, urllib) or access sensitive file paths. Operations are confined to mathematical calculations on input data provided via command-line arguments.
[Remote Code Execution] (SAFE): No use of eval(), exec(), or subprocess was found. The skill does not download or execute external code.
[Obfuscation] (SAFE): Code and documentation are clearly written and do not contain hidden characters, Base64 strings, or encoded commands.
[Indirect Prompt Injection] (SAFE): The skill accepts user-provided data (spacings and values) but strictly converts these inputs to floating-point numbers using float(x). This prevents any embedded text or instructions from being processed as strings by the agent.
[Unverifiable Dependencies] (SAFE): The skill uses only Python standard library modules (argparse, json, math, sys). No external packages are required.

convergence-study