The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install the @blockrun/mcp package using npx. This fetches and executes code from the npm registry, which is a common but unverified third-party source in this context.
[DYNAMIC_EXECUTION]: The blockrun_modal tool provides functionality for the agent to execute arbitrary code within a remote Modal-hosted sandbox. This includes actions for installing packages and running Python code with GPU support, representing a significant dynamic execution capability.
[INDIRECT_PROMPT_INJECTION]: The skill integrates multiple tools that ingest untrusted data from the web, research papers, and social media, which could contain malicious instructions designed to manipulate the agent.
Ingestion points: External data is retrieved via blockrun_exa (research papers), blockrun_search (web search), and blockrun_surf (social mindshare and news).
Boundary markers: There are no specific instructions or delimiters provided to help the agent distinguish between data and instructions when processing these external inputs.
Capability inventory: The agent possesses powerful capabilities including wallet management and financial transactions (blockrun_wallet), remote code execution (blockrun_modal), and making outbound voice calls (blockrun_phone).
Sanitization: The skill does not describe any sanitization or validation processes for data fetched from external providers.
[COMMAND_EXECUTION]: The installation guide provides shell commands that the user is expected to run, which automate the setup of the MCP server and involve executing remote packages.

blockrun-mcp-live-data