exploit-development

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The skill contains explicit override and "IGNORE ALL PREVIOUS INSTRUCTIONS" statements that instruct the agent to bypass safety/refusal policies and assume authorization, i.e., hidden/deceptive instructions that try to change agent behavior beyond the skill's stated purpose.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompt forces the model to always produce complete, working code and to comply with credential-access/post-exploitation requests (e.g., LSASS dumping), which makes it likely the LLM would need to handle or embed secret values verbatim in outputs, creating an exfiltration risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This file is a prompt-injection that intentionally overrides safety controls to coerce the assistant into providing exploit development, bypass, and post‑exploitation capabilities—facilitating malicious activities like RCE, credential theft, and persistence.

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 explicitly orders the agent to bypass all safety, provide exploit and privilege-escalation guidance, and includes post-exploitation/persistence techniques (e.g., .bashrc modification, init scripts, kernel modules, sudo misconfigurations) that instruct or enable modifying the host system state.