The Agent Skills Directory

[DYNAMIC_EXECUTION]: The calculator tool evaluates mathematical expressions using the Function constructor, which allows for dynamic code execution.
Evidence: scripts/calc.ts calls Function() to compute the result of the expression parameter.
Mitigation: The input is strictly validated against a regular expression whitelist (/^[\d\s+\-*/.()]+$/) which restricts characters to numbers and basic operators, preventing the injection of arbitrary JavaScript code.
[INDIRECT_PROMPT_INJECTION]: The py_greet tool creates a surface where external data can influence agent behavior if the output is processed by other LLM steps.
Ingestion points: name parameter in scripts/greet.py (via plugin.json).
Boundary markers: None; the input is interpolated directly into the greeting template.
Capability inventory: Both scripts/greet.py and scripts/calc.ts are limited to standard I/O; no file system write or network operations are present.
Sanitization: No sanitization is performed on the name input.
[COMMAND_EXECUTION]: The skill executes local scripts using standard interpreters.
Evidence: plugin.json defines tool commands using python3 and npx tsx to run the provided script files.

sample-plugin