The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to install the inbound-cli package globally using npm i -g. This command typically requires administrative privileges on many system configurations.
[COMMAND_EXECUTION]: The repository includes a security-agent package that provides an HTTP wrapper for executing administrative and abuse-handling commands (e.g., /abuse-block), as documented in its README.
[EXTERNAL_DOWNLOADS]: The root skill and the nested react-doctor skill instruct the agent to fetch and execute packages from the official NPM registry (inbound-cli and react-doctor). NPM is a well-known and trusted service registry.
[PROMPT_INJECTION]: The skill facilitates the processing of incoming email bodies, which are untrusted external data sources, creating a surface for indirect prompt injection (Category 8).
Ingestion points: Untrusted data enters the agent's context through incoming email bodies stored in the parsed_emails and structured_emails database tables.
Boundary markers (absent): The provided skill instructions do not specify the use of delimiters or instructions to ignore embedded commands within email content.
Capability inventory: The agent is empowered to list, retrieve, and modify email data using the inbound CLI and can potentially trigger actions through the security-agent and prompt-support modules.
Sanitization (absent): There is no explicit requirement in the skill body for the agent to sanitize or escape untrusted content before interpretation.

inbound-cli