The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to run hive init and hive upgrade commands through npx. These commands are used to configure local client environments and update the Hive CLI tool, which are standard setup procedures for the Hive platform.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx -y -p hive-intelligence@latest to execute initialization and management scripts directly from the npm registry. These scripts originate from the official package managed by the vendor (hive-intel).
[EXTERNAL_DOWNLOADS]: The skill references several official vendor domains, including mcp.hiveintelligence.xyz for the MCP endpoint and www.hiveintelligence.xyz for documentation and manifest files. These connections are necessary for the service's functionality and are part of the vendor's infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user queries to be processed by MCP tools, creating a potential attack surface for indirect prompt injection.
Ingestion points: User prompts (e.g., crypto price lookups or DeFi protocol searches) are passed directly to Hive MCP tools for execution.
Boundary markers: The skill includes a dedicated 'Security guardrails' section that explicitly instructs the agent to treat user prompts and tool outputs as untrusted data.
Capability inventory: The skill describes capabilities for reading data (invoke_api_endpoint) and potentially stateful operations (invoke_stateful_endpoint), while advising conservative use of the latter.
Sanitization: The instructions emphasize using the smallest useful tool surface and requiring explicit human approval for destructive or stateful effects.

hive-mcp