solana-dev
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches toolchains, CLI tools, and installation scripts from official Solana and Anza domains, which are recognized as well-known technology providers in this ecosystem.
- [REMOTE_CODE_EXECUTION]: Instructs the agent to configure and use a Model Context Protocol (MCP) server from an official Solana Foundation domain to provide live documentation and specialized expertise.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to interact with on-chain data which presents a known injection surface. This is properly addressed through instruction-level mitigations:
- Ingestion points: Interacts with account data, RPC responses, and program logs as detailed in
SKILL.md. - Boundary markers: Explicitly instructs the agent in
SKILL.mdto ignore any directives found in fetched data and to treat all on-chain content as untrusted. - Capability inventory: Uses
client.sendTransactionandclient.sendTransactionsfor blockchain interactions, and shell-based tool management as documented inreferences/kit/overview.md. - Sanitization: Mandates validation of RPC responses (ownership, discriminators) and transaction simulation before any execution requests.
Audit Metadata