casper

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The content is an explicit offensive authorization-bypass framework that automates credential abuse/authentication attempts, JWT manipulation (including "none" algorithm and signature/algorithm confusion), systematic data discovery/extraction, and financial/transaction manipulation (payloads directing funds to "attacker"), all of which enable deliberate unauthorized access, data exfiltration, and fraud — i.e., clear malicious exploitation capabilities rather than benign tooling.

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

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow performs HTTP requests to the operating user’s provided target URL and then ingests the returned response bodies as readable text (e.g., response=$(curl ... "$target_url$endpoint") and later echo "$response" | jq/grep/awk), so any outsider-controlled content returned by that target (public web content or attacker-controlled endpoints) can be fed into the agent’s LLM context via generated findings/report text.

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

• Direct money access detected (high risk: 1.00). The skill explicitly includes API examples and commands that perform/modify financial transactions (e.g., POST to /transfers or /transfer, "Test negative transfer", banking API assessment examples like https://bank.api.com/transfer and https://bank.api.com/transfers). It targets business-logic testing of financial transaction manipulation and demonstrates how to craft and send transaction requests to banking endpoints. Those examples constitute direct financial execution capability (sending transactions to bank APIs).