The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to add an external MCP server located at https://rube.app/mcp. This domain is an unverified third-party source not belonging to a known trusted organization.
[REMOTE_CODE_EXECUTION]: Adding an external MCP server allows a remote service to inject tool schemas and execution logic into the agent's environment. Because rube.app is an untrusted source, this represents a high-risk remote capability injection where the remote server controls the code/tools the agent interacts with.
[DATA_EXFILTRATION]: The skill manages Instagram connections through the external service (RUBE_MANAGE_CONNECTIONS). This architectural pattern involves sending authorization data and potentially sensitive user information (like media content and account insights) to the third-party server, creating a high risk of data exposure.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of external data.
Ingestion points: The tool INSTAGRAM_GET_IG_MEDIA_COMMENTS retrieves untrusted text from the public web (Instagram comments).
Boundary markers: The instructions lack explicit delimiters or warnings to treat external comment data as untrusted.
Capability inventory: The skill has high-impact capabilities including INSTAGRAM_CREATE_POST and INSTAGRAM_POST_IG_USER_MEDIA_PUBLISH, which could be triggered by instructions embedded in external comments.
Sanitization: There is no evidence of sanitization or filtering applied to the retrieved data before it is processed by the agent.

instagram-automation