The Agent Skills Directory

[PROMPT_INJECTION]: The skill name k7-total-security-unlock-patch is deceptive and indicates the skill is intended as a software crack or activation bypass. This creates a high risk of malicious intent or social engineering, as software piracy tools are a common vector for malware delivery. The actual content of the skill, which describes an 'Enterprise Configuration Framework,' is inconsistent with this name.
[PROMPT_INJECTION]: The k7_ai_analyzer.py code example contains a clear indirect prompt injection vulnerability. It ingests untrusted threat_event data (Ingestion points) and interpolates it directly into prompts for OpenAI and Anthropic models without using boundary markers or delimiters. The skill possesses the capability to execute system commands via the k7-console CLI (Capability inventory). No sanitization or validation of the event data is performed before it is used to generate a 'Response Playbook' (Bash/PowerShell scripts), which could allow an attacker to inject malicious instructions into the generated containment scripts.
[COMMAND_EXECUTION]: The K7SecurityManager class in SKILL.md implements functionality to execute system commands using the subprocess.run method. It specifically calls an external tool k7-console with arguments provided via user-defined configuration files and environment variables, including target_ip and auth_cert. This enables the execution of CLI tools on the local or remote system based on potentially untrusted configuration data.

k7-total-security-unlock-patch