The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the Solana Model Context Protocol (MCP) server from the official mcp.solana.com domain to provide the agent with access to live documentation and expert assistance.
[REMOTE_CODE_EXECUTION]: Downloads and executes the Solana/Agave toolchain installer from release.anza.xyz, which is the official distribution channel for Solana core development tools.
[REMOTE_CODE_EXECUTION]: Provides instructions to install the Surfpool development utility via a shell script from run.surfpool.run (or its GitHub releases), a well-known tool in the Solana ecosystem for local networking and testing.
[COMMAND_EXECUTION]: Invokes various development CLI tools, including anchor, surfpool, and cargo, using the NO_DNA=1 environment variable to ensure non-interactive and structured output suitable for AI agents.
[DATA_EXPOSURE]: The skill explicitly defines an 'untrusted data handling' policy for data ingested from the Solana blockchain. It requires the agent to validate account owners and data length, and to ignore any instructions embedded in on-chain data (e.g., in metadata or logs) to prevent indirect prompt injection.
[CREDENTIALS_SAFE]: Contains strict instructions prohibiting the agent from requesting, generating, or storing private keys or seed phrases, ensuring that all signing operations are delegated to secure, user-approved wallet flows.

solana-dev