The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script using powershell.exe to perform the validation logic.
[DATA_EXFILTRATION]: The scripts read XML files from the filesystem and allow writing validation reports to a specified path. This is standard functionality for a validation tool and does not involve network exfiltration.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and displays text content (e.g., synonyms, comments) from untrusted XML files directly into the agent's output stream without sanitization.
Ingestion points: Content is read from user-provided XML files in subsystem-validate.ps1 and subsystem-validate.py.
Boundary markers: The output uses simple headers (e.g., === Validation: Subsystem.Name ===) but lacks robust delimiters to prevent the agent from interpreting file content as instructions.
Capability inventory: The skill can read and write files via the agent's tools.
Sanitization: No filtering or escaping of the XML text values is performed before display.
[SAFE]: The homoglyph detection for '1С' is a false positive; 'С' is the Cyrillic character used in the official name of the 1C:Enterprise platform, which this skill is designed to support.

subsystem-validate