okx-agentic-wallet

Fail

Audited by Snyk on Nov 16, 2026

Risk Level: CRITICAL
Full Analysis

CRITICAL E004: Prompt injection detected in skill instructions.

  • Potential prompt injection detected (medium risk: 0.30). The skill is a legitimate OKX wallet, but Gas Station's Scene A and Scene C have the agent change the user's default gas-token setting via update-default-token without showing that command, which hides a settings change from the user, though it is echoed afterward and follows the user's chosen token.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

  • Third-party content exposure detected (medium risk: 0.30). The required wallet balance and public-address portfolio flows read symbol/tokenName metadata for every token in a wallet, and anyone can airdrop a token with attacker-chosen text into a monitored wallet without the agent selecting it first.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

  • Direct money access detected (medium risk: 0.30). The skill operates OKX wallets and performs direct financial operations: sending and transferring tokens, swaps, bridges, limit orders, contract calls, and broadcasting transactions (SKILL.md lines 29-30, 36-39). These are direct operations, not indirect ones. For control, the skill documents a wallet login flow (line 25) and backend/server-side checks. Examples are the backend's "confirming" responses, server validation of balances and gas, and email-code approval verification for token approvals (lines 81-87). Wallet access requires authentication (login with credential store, session tokens, TEE signing), and the service layer enforces it. Credential-corruption errors force re-login (line 94), which shows that the service or execution layer gates access. The confirmation prompts are agent-followed and would not count alone. The authenticated wallet session and the backend approval verification (email code) are service-enforced access controls, though, so the direct operations have qualifying independent access control. That places this at medium rather than high.

Issues (3)

E004
CRITICAL

Prompt injection detected in skill instructions.

W011
MEDIUM

Third-party content exposure detected (indirect prompt injection risk).

W009
MEDIUM

Direct money access capability detected (payment gateways, crypto, banking).

Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 16, 2026, 11:43 AM
Issues
3