The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Installs the @intentsolutionsio/ccpi package globally from the NPM registry using pnpm.
[REMOTE_CODE_EXECUTION]: Instructs the agent to add a third-party GitHub repository (jeremylongshore/claude-code-plugins) as a plugin marketplace and install community-contributed plugins, which involves executing remote scripts or prompt instructions.
[COMMAND_EXECUTION]: Executes system-level commands via the Bash tool to perform global package installations and marketplace operations.
[INDIRECT_PROMPT_INJECTION]: The skill creates a significant attack surface by ingesting and executing instructions from community-contributed plugins.
Ingestion points: Plugin content downloaded via the ccpi tool or added through the /plugin marketplace command.
Boundary markers: No explicit boundary markers or isolation instructions are provided for the content of the installed plugins.
Capability inventory: The agent retains Bash, Read, Write, and WebFetch capabilities, which can be exploited if a malicious plugin is installed.
Sanitization: There is no evidence of validation or sanitization for the plugin metadata or executable content before installation.

ccpi-marketplace