The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation provides instructions to download and execute an installation script from the vendor's public repository.
Evidence: examples/mcp-server/README.md contains the command curl -fsSL https://raw.githubusercontent.com/sendaifun/solana-mcp/main/scripts/install.sh -o solana-mcp-install.sh which fetches an external shell script.
[REMOTE_CODE_EXECUTION]: The guide instructs users to execute downloaded scripts and run remote packages via npx, which facilitates the execution of code from the vendor's remote sources.
Evidence: examples/mcp-server/README.md includes ./solana-mcp-install.sh and npx solana-mcp as standard setup steps.
[COMMAND_EXECUTION]: The skill scripts and configuration files enable the AI agent to execute shell commands and interact with the local operating system to manage global NPM packages and MCP server processes.
Evidence: examples/mcp-server/README.md provides configuration for Claude Desktop to run npx solana-mcp or direct node execution of the server package.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an environment where agents process untrusted user input to perform high-impact blockchain operations, creating a surface for potential prompt injection attacks.
Ingestion points: User input is ingested via standard input streams in templates/agent-template.ts and examples/langchain/README.md, as well as via API routes in examples/vercel-ai/README.md.
Boundary markers: The system prompts include specific guidelines such as "Never reveal private keys" and instructions to confirm transactions over 0.5 SOL.
Capability inventory: The toolkit provides over 60 actions, including token transfers, swaps via Jupiter, NFT minting, and airdrops, all performed via network RPC calls.
Sanitization: There is no evidence of explicit input sanitization or validation of user-provided natural language before it is processed by the language model to determine tool usage.

solana-agent-kit