The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill depends on an external MCP server ('threenix') to provide component source files, integration guidelines, and dependency lists during execution.
[REMOTE_CODE_EXECUTION]: Content retrieved from the 'threenix' tool is written to the local project and subsequently processed by build or type-checking tools, enabling the potential execution of remote-sourced code in the local environment.
[DYNAMIC_EXECUTION]: The skill generates local source files (via UTF-8 or Base64 decoding) from strings provided by the external tool, which prevents static verification of the code being introduced into the project.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from tool outputs that could contain malicious instructions to subvert agent logic during file writing or package installation.
Ingestion points: The files[] and integrationNotes fields returned by the get_component_reference tool in SKILL.md.
Boundary markers: None; the instructions require the agent to write the returned content verbatim.
Capability inventory: File system write access, automated package installation, and shell command execution for builds/type-checks.
Sanitization: The skill does not define any validation or sanitization routines for the content received from the remote MCP server.

add-fireworks