The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill describes an execution environment where the agent generates and runs Python code cells (execute_python). It details the use of RestrictionsConfig to block hazardous modules (e.g., subprocess, socket) and functions (e.g., exec, eval, os.system). These are intended architectural features with documented security mitigations.
[INDIRECT_PROMPT_INJECTION]: The skill details how the agent ingests parameters and external context which are then used within the code execution loop. It documents boundary markers and sanitization strategies, such as spec() parameter constraints and TruncationConfig limits. * Ingestion points: Agent method parameters and data processed within execute_python cells. * Boundary markers: spec() annotations and TruncationConfig character caps. * Capability inventory: Python REPL execution (execute_python), media viewing (show), and result validation (return_result). * Sanitization: RestrictionsConfig denylist and AST-based validation.

nooa-codeact-advanced