building-agents-construction

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The prompt includes authoritative, concealment and override-style directives (e.g., "DO NOT DISPLAY THIS FILE", "IMMEDIATELY begin executing... Do not explain what you will do") that instruct the agent to hide instructions and perform actions without disclosure or explicit consent, which is deceptive and outside ordinary transparent behavior for a skill description.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill contains strong prompt-injection instructions to immediately execute hidden build steps, spawn a local MCP server process, and write/execute package files—which enable remote code execution, filesystem modification, and a covert backdoor path despite no explicit data-exfiltration strings being present.