The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions to remain silent after calling certain tools are identified as user experience optimizations. By relying on the platform's real-time task cards and progress checklists to inform the user, the skill avoids redundant status messages without concealing the underlying agent activities.\n- [COMMAND_EXECUTION]: The skill orchestrates complex task lifecycles using internal platform tools like task-control, workflows, and executions. It provides specific logic for error handling, task re-planning, and workflow execution based on explicit user requests or system follow-up events.\n- [SAFE]: The skill demonstrates defensive engineering by instructing the agent to 'never trust builder prose' and requiring manual inspection of workflow JSON graphs. This ensures that verification outcomes are based on actual code structure rather than potentially unreliable natural language summaries.

planned-task-runtime