juicebox-v6
Audited by Snyk on Nov 17, 2026
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (medium risk: 0.30). The Bendystraw GraphQL indexer, which the jb-bendystraw module queries for project payment history and activity feeds, returns
payEvent.memoand project metadata that any payer can write into a project's feed 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 operational path is the catalog's instructed install of the skill from the GitHub repo
BankrBot/skills, which is an organization-owned distribution channel that is not clearly first-party for this upstream content, so it is classed as common/reputable unofficial and rated medium (the frontend npm tarballs are pinned by integrity and are official, and the Bendystraw, Relayr, and RPC URLs are data or control-plane, not inbound dependencies).
MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).
- Direct money access detected (high risk: 0.70). The skill covers operational on-chain financial actions: paying projects, cash-outs, payouts, REVLoans operation, and bridging. It builds transaction plans and, via Bankr's wallet or the user's signer, can sign and broadcast them (lines 3, 18, 20, 29, 33, 38). These are direct operations (payments, cash-outs, loans, swaps/transfers), not just development guidance or reading.
Controls: the skill documents many safeguards, but none is an independent authentication or authorization check enforced by a service or execution layer. The safeguards are:
- a deployment manifest check (13)
- a pipeline of preview, simulation, and verification (15-17)
- explicit user confirmation (18)
- a slippage minimum (16)
- post-transaction receipt checks (19)
- keeping private keys out of code (20)
These are agent-followed instructions, user confirmation, and local signing. The signer flow (Bankr or the user's approved signer) is not described as an access-control mechanism for the payment route. Under the criteria, agent-followed confirmations and local signing do not qualify as independent access validation. Direct operations without an established qualifying control are High.
Issues (3)
Third-party content exposure detected (indirect prompt injection risk).
Unverifiable external dependency detected (runtime URL that controls agent).
Direct money access capability detected (payment gateways, crypto, banking).