The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local PowerShell script (scripts/meta-compile.ps1) to perform metadata generation. Automated analysis of the skill's associated scripts (specifically scripts/meta-compile.py) has identified the use of subprocess.call() for executing shell commands. This implementation can lead to command injection if the arguments provided to the script, such as output directory paths or JSON content, are not rigorously sanitized before being processed by the shell.
[DYNAMIC_EXECUTION]: The helper script scripts/meta-compile.py uses subprocess.call() to run commands at runtime. This dynamic execution of shell strings is a known risk vector that could allow for unauthorized system operations if the script's inputs are manipulated by an attacker.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading data from a JSON file to determine what metadata to create and where to write files, creating an attack surface where maliciously crafted JSON could potentially exploit the skill's file-writing capabilities. 1. Ingestion points: JSON file paths provided via the -JsonPath argument in SKILL.md. 2. Boundary markers: No delimiters or instructions to ignore embedded commands are documented for the JSON input processing. 3. Capability inventory: The skill performs subprocess calls (scripts/meta-compile.py) and file system writes/modifications within the specified output directory and Configuration.xml (SKILL.md). 4. Sanitization: The provided documentation does not indicate any validation or sanitization logic to prevent malicious payloads within the JSON or path arguments.

meta-compile