The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is primarily designed to generate and execute full Python scripts at runtime via the pyats_run_dynamic_test tool. It accepts a test_script parameter containing source code as a serialized JSON string, which is then executed in a Python environment. While the skill documentation explicitly bans several dangerous modules (e.g., os, subprocess, requests) and functions (e.g., eval, exec, open), the reliance on runtime code generation and execution is a significant security surface.
[COMMAND_EXECUTION]: The execution mechanism involves invoking a shell command that passes the dynamic Python script to a management tool. This pattern allows for user-influenced data (via the agent's generation of the test_script) to be executed within the environment.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect injection where an attacker could influence the data the agent processes (such as device outputs or external documentation) to cause the agent to generate a test_script with unintended or malicious logic.
Ingestion points: The test_script parameter of the pyats_run_dynamic_test tool, which is populated by the agent based on its current context.
Boundary markers: There are no explicit boundary markers or isolation instructions for the data used to populate the TEST_DATA dictionary within the script.
Capability inventory: The skill possesses the capability to execute code, connect to network devices via a testbed, and log results.
Sanitization: The instructions list prohibited Python imports and functions as a mitigation strategy, but effectiveness depends on the runtime enforcement by the pyats_run_dynamic_test implementation rather than the agent's instructions.

pyats-dynamic-test