Back to skill

Security audit

Arsenal — Sumplus DeFi Execution Layer

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed DeFi integration, but it can route users into high-impact blockchain transactions from remote-generated payloads without enough validation safeguards.

Install only if you trust Arsenal/Sumplus for DeFi execution and are prepared to manually verify every transaction in your wallet. Do not sign opaque transactions based only on the skill's description or quote; check recipients, approvals, token amounts, chain, contract addresses, and any persistent allowances. Treat API key creation as an explicit account-management action, not something an agent should do silently.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:69
Finding

Unvalidated Remote Blockchain Transactions May Cause Unauthorized Asset Transfers

Content
View full analysis
", "instructions": ["Pass tx_bytes to any Sui wallet to sign and submit."] } ``` Pass `tx_bytes` to Privy `send_sui_transaction` or the user's Sui wallet. ### Solana transaction (`build_swap_tx`) ```json { "transaction": "", "amount_in": "0.1", "amount_out": "14.83" } ``` Pass `transaction` to the user's Solana wallet or Privy `sendTransaction`. ``` ### Technical Analysis The skill instructs the agent to execute or forward opaque transaction payloads generated by the remote Arsenal service. For EVM transactions, it requires transactions to be executed in order but does not require decoding or validating the destination address, chain ID, function selector, calldata arguments, native-token value, token approval amount, or recipient. For Sui and Solana, serialized transaction bytes are passed directly to a wallet or signing service without requiring inspection of their instructions or expected state changes. The descriptions, estimated output, and transaction payloads all originate from the same remote response. Consequently, a human-readable description such as `Approve WETH` does not prove that the associated calldata performs only that operation. Confirmation of an earlier quote also does not establish that a subse ...[truncated 2190 chars]
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description says to use the skill for 'any blockchain operation,' which is too expansive for safe routing. Because the skill can prepare transactions across many chains and protocols, broad matching materially increases the risk of accidental activation in sensitive financial contexts.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction to use Arsenal for 'anything blockchain-related' is overly broad and can cause the agent to invoke the skill on loosely related prompts without sufficient scoping. In a high-risk domain like DeFi, over-triggering increases the chance of unnecessary data disclosure, unintended transaction preparation, or execution pathways being reached too easily.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 12)May include surrounding context.

md
Arsenal is Sumplus's on-chain execution platform. When the user asks for anything blockchain-related — swaps, lending, balances, liquidity positions — call Arsenal.

**Base URL:** `https://arsenal.sumplus.xyz`
**Auth:** `Authorization: Bearer {ARSENAL_API_KEY}` (inject from env, never ask the user)

---

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill claims to be for DeFi and blockchain operations, but it also documents account signup and API key creation endpoints. That expands the capability beyond the declared purpose and could let an agent create new service accounts or provision credentials without a clear user request or authorization boundary.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Including flows to register Arsenal accounts and mint API keys is a privileged identity-management capability that is unrelated to ordinary quote/transaction building. In an agent context, this can be abused to create shadow accounts, generate fresh secrets, and bypass intended administrative controls.

Content

No source excerpt is available for this finding.