The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill prompts for the global installation of an external package via npm install -g packx. This introduces a dependency on a third-party package that is not part of the trusted sources list, posing a potential supply chain risk.
COMMAND_EXECUTION (LOW): The skill is centered around executing the packx command-line utility with various flags and shell redirection.
SYSTEM_LIMIT_BYPASS (LOW): The documentation explicitly instructs the agent to use stdout > file.md instead of the tool's built-in output flag to 'avoid triggering WriteFile hook size limits'. This is a deliberate instruction to bypass environment-level security or monitoring hooks intended to regulate file operations.
INDIRECT_PROMPT_INJECTION (LOW): The skill's primary purpose is to ingest untrusted data (source code) into the agent's context, creating a surface for indirect prompt injection.
Ingestion points: Any local file path provided to the tool (e.g., src/).
Boundary markers: The tool uses XML format by default, but the skill lacks explicit 'ignore embedded instructions' warnings for the agent when processing the resulting bundle.
Capability inventory: File system read access, clipboard write access (-c), and shell redirection.
Sanitization: No sanitization of the bundled content is performed before it is presented to the agent.

packx