nexus

Fail

Audited by Snyk on Nov 17, 2026

Risk Level: CRITICAL
Full Analysis

CRITICAL E004: Prompt injection detected in skill instructions.

  • Potential prompt injection detected (medium risk: 0.30). The skill's main purpose is legitimate, but it tells the agent to help users create Bankr keys without the allowedRecipients safety restriction and promotes the $NEXUS token, which weakens user protections and inserts unrelated crypto promotion, without showing malicious intent.

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

  • Third-party content exposure detected (medium risk: 0.30). The required copy-trade and top-caller workflows read the public /feed and /theses/leaderboard, which include outsider-written notes and displayName fields that anyone can publish (including by calling the public registry contract), so outsiders can feed text into a source the workflow consumes without the agent selecting a specific item.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

  • Potentially malicious external URL detected (medium risk: 0.30). The only inbound source is the catalog's instructed install of the skill from a GitHub repository under the BankrBot organization, a reputable project org without strong first-party proof for the Nexus content, so it is a common/reputable unofficial dependency (medium); the Nexus API, Orderly, RSS, and Arbiscan URLs are outbound data, status, or inspection references and fail the dependency gate.

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

  • Direct money access detected (medium risk: 0.30). Direct operations are established: /trade places leveraged perpetual orders that move real collateral (SKILL.md:59-67), /proxy/bankr-withdraw withdraws USDC (line 251), and the autonomous agent trades real funds once live (lines 78-81, 98, 106-108). These are direct, not indirect. Control is documented for these routes. Every authenticated endpoint requires a walletSig (lines 19-28, 241, 251). For agent mutations the server ecrecovers the signature and rejects unless it resolves to the agent's own wallet (lines 113-116). The signature check is therefore enforced by the service, not by the agent following instructions. The Live activation gate and confirm: "GO LIVE" are agent-followed or confirmation flags and are not counted as the control. The webhook path is PAPER by default and its bearer token is not treated as the qualifying control. The qualifying control is the service-enforced wallet-signature check, so the tier is Medium rather than High.

Issues (4)

E004
CRITICAL

Prompt injection detected in skill instructions.

W011
MEDIUM

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

W012
MEDIUM

Unverifiable external dependency detected (runtime URL that controls agent).

W009
MEDIUM

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

Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 17, 2026, 07:41 AM
Issues
4