The Agent Skills Directory

[SAFE]: The skill defines a structured orchestrator-worker pattern using the environment's native Task tool for local concurrency.
[PROMPT_INJECTION]: The skill provides robust guidelines for managing subagent 'contracts' and specifically addresses the risk of indirect prompt injection. It instructs the agent to treat subagent outputs as unverified claims and provides a dedicated 'Verification' framework (e.g., verifier workers, evidence checking, and source cross-referencing) to ensure integrity.
[COMMAND_EXECUTION]: The skill mentions using standard utilities like ssh, rsync, and curl for administrative tasks such as data staging (pulling remote data to local environments) and validating final deliverables. These operations are described within the context of legitimate developer and research workflows.
[EXTERNAL_DOWNLOADS]: The documentation references well-known and reputable technical sources including Anthropic, OpenRouter, and various academic papers (arXiv). These references are provided for methodological context and do not involve the execution of untrusted remote code.

parallel-orchestrate