The Agent Skills Directory

[SAFE]: The skill establishes a process for AI-driven code review and model comparison using standard delegation tools. It focuses on orchestration, reporting, and adversarial checking.
[REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution. The skill relies on platform-defined delegation tools (delegate_task) to communicate with other models within the agent's supported infrastructure.
[PROMPT_INJECTION]: The instructions do not attempt to bypass safety filters or override system-level instructions. Instead, they reinforce constraints such as maintaining fresh context, preserving safety requirements, and adhering to existing review contracts.
[CREDENTIALS_UNSAFE]: No hardcoded API keys or sensitive secrets were found. The model selection and provider routing are handled via standard configuration keys.
[DATA_EXFILTRATION]: The skill is focused on internal processing and reporting within the agent's chat or authorized PR descriptions. It does not include network operations to external, non-whitelisted domains for data harvesting.
[PRIVILEGE_ESCALATION]: The skill explicitly restricts its own authority, stating it 'grants no comment, commit, push, PR update or merge permission' and must use existing shipping contracts.

council