The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, including Python source files for static analysis and mathematical expressions for calculations.
Ingestion points: audit_units.py ingests local Python files, while propagate_uncertainty.py and check_plausibility.py accept mathematical expressions as CLI arguments.
Boundary markers: Mathematical expressions are parsed into an Abstract Syntax Tree (AST) to separate logic from data.
Capability inventory: The bundled tools have the ability to read local files and write reports to the filesystem. No network capabilities or arbitrary command execution mechanisms were identified.
Sanitization: The skill employs a secure 'sandbox' approach for expression evaluation; it parses input into an AST and manually reduces the tree using a strict whitelist of allowed nodes and functions, successfully avoiding the security risks associated with eval() or exec().

uncertainty-and-units