The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The MCP server installation guide in examples/mcp-server/README.md instructs users to download a shell script from an external repository using curl -fsSL https://raw.githubusercontent.com/sendaifun/solana-mcp/main/scripts/install.sh.
[REMOTE_CODE_EXECUTION]: The skill provides instructions to grant execution permissions (chmod +x) to the downloaded remote script and execute it immediately (./solana-mcp-install.sh). This pattern allows for arbitrary code execution from a third-party source during the setup process.
[COMMAND_EXECUTION]: The toolkit frequently uses shell-based commands for setup and provides interactive CLI templates (templates/agent-template.ts) that execute blockchain logic via ts-node based on user input.
[PROMPT_INJECTION]: The skill creates a high-risk surface for indirect prompt injection by exposing tools that perform irreversible financial actions (e.g., transfer, trade, lend) based on raw natural language input.
Ingestion points: User-provided strings in chat interfaces, autonomous task descriptions, and MCP tool inputs.
Boundary markers: The skill includes system prompt instructions such as "Never reveal private keys" and "Always confirm large transactions," which provide some protection but do not eliminate the underlying injection risk.
Capability inventory: The agent has broad capabilities including token transfers, DeFi pool creation, and automated trading across all integrated scripts.
Sanitization: There is no evidence of input validation or filtering before passing user instructions to the LLM for tool selection and execution.

solana-agent-kit