[agentbox] security: XAPS pre-execution audit before on-chain actions

[SentVan23]

Context
cascade-protocol/agentbox runs agent-controlled value movement in packages/openclaw-x402/src/index.ts. A lightweight pre-execution audit (~200–300ms) before signing or sending can catch prompt-injection-driven tx mistakes without blocking normal flow.

Matched code (packages/openclaw-x402/src/index.ts)

.option("-o, --output <dir>", "Output directory for wallet files")
        .action((opts: { output?: string }) => {
          if (!opts.output) {
            console.error("Usage: openclaw x40

XAPS is a pre-execution audit tollbooth for agents: dual-LLM + rules, ~$0.01 per approved audit, returns a portable receipt. Not a walled garden — a guardrail on the open road.

Suggested integration

from xaps import XapsClient, XapsRejectedError

receipt = XapsClient(base_url="https://api.xaps.network").audit(
    action="execute_smart_contract",
    contract_address=TARGET_CONTRACT,
    amount=float(AMOUNT),
)
if receipt["audit"]["status"] != "APPROVED":
    raise XapsRejectedError(receipt["audit"].get("beta_attack", "REJECTED"), receipt=receipt)

Next steps

1. Drop the snippet at the hot path (or add MCP io.github.APMC1/xaps).
2. Register an agent wallet: POST https://api.xaps.network/agents/register ($1 trial).
3. Agent card + MCP install: https://api.xaps.network/.well-known/xaps-agent-card.json
4. Happy to open a PR with the exact insertion point if useful.

— XAPS team (distribution outreach; replies welcome)