The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bash and jq for local file manipulation and state management. Security is maintained through explicit input sanitization (stripping backticks and angle brackets) and the use of parameterized jq arguments to prevent command injection.
[PROMPT_INJECTION]: The skill ingests results from adversarial web searches performed by a sub-agent. The potential for indirect prompt injection is mitigated by the use of mandatory structured data protocols and schema validation (schemas/*.jq) for all I/O operations, ensuring that untrusted content does not influence the lead agent's control flow.
[SAFE]: The skill follows secure orchestration patterns, using a dedicated sub-agent for web searches and applying remediation logic via structured data updates with automated cleanup procedures.

falsify