The Agent Skills Directory

[SAFE]: The skill demonstrates safe orchestration patterns, utilizing a designated _workspace/ directory for intermediate data and strictly referencing local documentation for its logic. It does not perform network operations, execute arbitrary code, or access sensitive system paths.
[PROMPT_INJECTION]: As an NLP orchestrator, the skill processes untrusted user input that is passed to sub-agents. This constitutes a standard surface for indirect prompt injection. While the risk is low, the orchestrator instructions do not specify explicit delimiters for the processed text.
Ingestion points: _workspace/{run_id}/01_input.txt (Phase 1).
Boundary markers: Absent; file paths are passed to agents without explicit delimiters.
Capability inventory: The skill uses Agent and TeamCreate tools and performs local file read/write operations.
Sanitization: No explicit sanitization or filtering is mentioned.

humanize-korean