llm-prompt-injection

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The playbook contains explicit examples that tell the LLM to read sensitive data (system prompt, config files, conversation context) and embed it verbatim into HTTP requests, tool arguments, or outputs, which requires the model to handle/output secrets directly.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The documents are an explicit adversary playbook describing concrete prompt-injection, tool-abuse, data-exfiltration, RCE, and system-prompt-extraction techniques that can be used to create backdoors and steal sensitive data.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill includes explicit examples that instruct reading sensitive system files (e.g., /etc/passwd), executing arbitrary shell commands (e.g., curl ... | bash), and chaining tool calls to exfiltrate data — all of which can lead to RCE or system compromise if executed by an agent with tool/shell access.