The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages a permanent authentication token (anon_...) used to access the VoiceDrop service. The instructions explicitly state that this token is irrevocable and does not expire. Furthermore, it warns that the token will be stored within the model context and conversation history, which creates a potential risk for data exposure if session logs are accessed.
[COMMAND_EXECUTION]: The skill provides shell command snippets (e.g., curl -s -X POST ... | python3 -c ...) for interacting with the service API. These scripts are intended to be executed by the agent or user to perform login and file upload operations.
[EXTERNAL_DOWNLOADS]: The skill communicates with external endpoints at voicedrop.cn and jianshuo.dev. While these domains are associated with the vendor, they involve sending data (including potential session tokens and binary files) to remote servers.
[REMOTE_CODE_EXECUTION]: The skill provides a mechanism to interact with a remote Model Context Protocol (MCP) server. If the remote server provides executable instructions or data that the agent processes without sanitization, it could lead to unintended behavior.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through tools that read external content (e.g., list_articles, community_feed). There are no boundary markers or explicit sanitization instructions to prevent instructions embedded in that data from influencing the agent.
Ingestion points: Data returned by MCP tools like list_articles and read_style from voicedrop.cn.
Boundary markers: Absent; the instructions do not specify delimiters for external content.
Capability inventory: The skill uses curl for network operations and performs file uploads to jianshuo.dev.
Sanitization: None; the skill lacks logic to escape or validate incoming external data before it enters the prompt context.

wjs-voicedrop