themonitor
Audited by Snyk on Nov 17, 2026
CRITICAL E004: Prompt injection detected in skill instructions.
- Potential prompt injection detected (medium risk: 0.30). The skill's purpose is disclosed and its permissions are scoped and checked, but it asks the user to turn off Bankr's permitted-recipients safeguard and gives an unexplained instruction to avoid the
/api/live/feeendpoint, which is a sketchy weakening and concealment-like pattern without clear malicious intent.
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (medium risk: 0.30). Step 6 requires the agent to call
POST /api/live/stateand readpositions[].symandtrades[].sym, andsellallreturnssold[].sym; these token symbols are set by whoever deploys a token on Elysium, and automated leader-following can place such tokens into the account without the agent selecting them. Because the trigger is partly outside an outsider's control, I am scoring this as medium under the recall-first rule, though the attacker cannot directly force delivery.
MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).
- Direct money access detected (medium risk: 0.30). Direct financial operations are established: the skill deposits native HYPE from the Bankr wallet to the smart account (Step 4, lines 124-127), buys and sells tokens through the guarded swap contract (Step 5 and the
swappermission, lines 22, 137), sells positions and adds funds (Step 7, lines 155-156), and withdraws HYPE and pays fees (Step 8, lines 166-168). These are transfers, trades and withdrawals, not indirect operations.
Access control is documented and enforced by the service and execution layer, so this is Medium rather than High. The Elysium server requires a bearer token from a wallet-signed nonce/verify flow for the live routes, and returns 401 without it (lines 80-82, 48, 71, 189). The agent submits the signature, but the server decides whether the token is issued and whether the live routes accept the request. The on-chain permissions also require the user's own EIP-712 signature, and the server-side checks restrict each permission to the documented route (lines 21-25, 105-111). Login and session authentication gating trading and withdrawal count as qualifying control under the rubric. The agent-run checks in check-permission.mjs and the agent-followed confirmations are not counted as the control.
Because the answer is Medium rather than High, the locations cite the operations and the established access-control evidence.
Issues (3)
Prompt injection detected in skill instructions.
Third-party content exposure detected (indirect prompt injection risk).
Direct money access capability detected (payment gateways, crypto, banking).