The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill suggests an alternative configuration that executes a third-party package using npx -y @porkll/siyuan-mcp. This pattern downloads and runs code from a remote registry that is not associated with a verified vendor or the skill author.
[COMMAND_EXECUTION]: The skill utilizes shell commands like curl and jq to interact with the SiYuan API. These commands incorporate environment variables and identifiers that could be manipulated if not handled carefully by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by retrieving data from an external knowledge base.
Ingestion points: Data enters the agent's context through the /api/block/getBlockKramdown, /api/query/sql, and /api/search/fullTextSearchBlock endpoints.
Boundary markers: The instructions do not define clear delimiters or warning markers to prevent the agent from following instructions potentially embedded in the retrieved notes.
Capability inventory: The agent is given capabilities to create, modify, and delete documents and blocks, which could be abused if malicious instructions are injected into the knowledge base.
Sanitization: No sanitization or validation of the Markdown content returned by the API is described in the instructions.

siyuan