Proposal: PaySafe action provider (x402 payment-security firewall)

[corbinallison]

AgentKit agents that pay over x402 have no built-in guard against the failure modes that drain agent wallets — prompt-injection-triggered payments, replayed nonces, overpayment, secrets leaking in payment metadata, lookalike-token contracts, address poisoning. PaySafe (https://paysafe-agent.com) is a non-custodial firewall that returns an allow/flag/block verdict for a payment before it settles.

I maintain agentkit-paysafe (published, tested) and would like to contribute it in-tree as an action provider (action_providers/paysafe/) — three actions: paysafe_scan_payment, paysafe_check_reputation, paysafe_report_counterparty. It's already built against ActionProvider / @create_action and is network-agnostic (advisory metadata analysis; never touches the wallet).

Before I open the PR: are you open to a third-party service integration in-tree, or would you prefer it stay an external package? Happy to follow whatever pattern you point me to.

[doteyeso-ops]

Agree the failure modes you list (injection-triggered pay, replayed nonce, overpay, metadata leak, lookalike asset) are the ones that empty agent wallets — a pre-settle allow/flag/block is the right shape.

Complementary layer we run hosted (Solana USDC x402): preflight spend / idempotency / state guards + signed action-receipt after the hop, so you get both “should this pay?” and “did this settle with a verifiable receipt.” Prepaid key path so the agent isn’t re-signing every $0.02 check: https://vibes-coded.com/start

Not a substitute for PaySafe’s wallet firewall — different machine (policy gate + provenance vs custody firewall). Useful as a second surface if AgentKit wants both verdict-before-settle and offline-verifiable settle proof.