The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically generates batch files (.cmd) in the user's temporary directory (%TEMP%) to serve as launchers for background processes. These scripts are subsequently executed using the Win32_Process.Create WMI method to ensure the processes are detached and not terminated by the agent's shell timeout.
[PRIVILEGE_ESCALATION]: The commands generated to initiate background tasks include the -ExecutionPolicy Bypass flag. This is used to circumvent local system security policies that may otherwise restrict the execution of PowerShell scripts.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing untrusted data from the local environment and network.
Ingestion points: The script wait-ready.ps1 performs network requests to user-specified HealthUrl endpoints, and classify-failure.ps1 reads application stderr logs from the file system.
Boundary markers: There are no explicit boundary markers or sanitization procedures for the data retrieved from HTTP responses or log files before they are presented to the agent or used for decision-making.
Capability inventory: The skill has high-privilege capabilities, including process termination via taskkill.exe, detached process creation via WMI, and arbitrary file system writes within the state directory.
Sanitization: While the skill uses ProcessStartInfo.ArgumentList to mitigate direct shell injection when starting the target application, it remains susceptible to malicious instructions embedded in the external data it ingests.

long-running-process