The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses an unencrypted HTTP endpoint (http://mcp.bmobapp.com/mcp) for its MCP server. This causes sensitive credentials, specifically X-Bmob-Application-Id and X-Bmob-REST-API-Key, to be transmitted in plain text, making them vulnerable to interception on the network.
[COMMAND_EXECUTION]: The deploy_static_site tool instructs the agent to immediately execute shell commands provided by the remote MCP server (e.g., curl --data-binary @filePath). This creates a pattern where the agent acts as a remote command executor for the server's output.
[EXTERNAL_DOWNLOADS]: The skill facilitates the synchronization and deployment of cloud function source code from external endpoints (api.codenow.cn). While these appear to be vendor-owned resources, they involve the transfer and potential execution of external code logic.
[INDIRECT_PROMPT_INJECTION]: The skill has an injection surface by ingesting untrusted data from the remote backend (schema definitions and cloud function source code).
Ingestion points: Data enters the context via get_project_tables, list_cloud_functions, and get_cloud_function (SKILL.md).
Boundary markers: None identified to separate external data from agent instructions.
Capability inventory: The skill has the capability to write files to the local disk (get_cloud_function) and execute shell commands (deploy_static_site).
Sanitization: There is no evidence of sanitization for the data retrieved from the remote server before it is used by the agent.

bmob-mcp