The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches schema and verification guidelines from the author's GitHub repository (csark0812/toolbox). These resources are used as technical references for structuring subagent outputs and performing readiness checks.
[COMMAND_EXECUTION]: Orchestrates the host's native Task (or Subagent) tool to spawn parallel workers. The orchestration kernel includes strict functional constraints to ensure sub-tasks are executed according to the plan and synthesized upon completion rather than skipped.
[PROMPT_INJECTION]: The skill implements an "adversarial" debate mode using specialized stances (e.g., attacker, refuter, premises) to challenge claims and improve output quality. These are structured prompting techniques for analysis and do not attempt to bypass AI safety filters.
[SAFE]: The skill manages agent discovery by parsing metadata from local configuration files (.claude/agents/*.md) and implements "context asymmetry" to prevent subagents from being influenced by the parent chat's internal reasoning, which serves as a security best practice for multi-agent orchestration.

multi