The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The orchestration logic involves passing untrusted user requirements and external data (e.g., UI sources fetched via URL or MCP) across multiple specialized subagents. While the skill defines clear handoff contracts (HC-01 through HC-06), it lacks explicit instructions for subagents to use boundary markers or sanitization when processing these external inputs.
[COMMAND_EXECUTION]: The orchestrator is authorized to use the Bash tool for repository operations and verification. Subagents such as task-executor and quality-fixer are delegated authority to modify the codebase and execute quality checks, which is standard for autonomous development agents but remains a capability to be monitored.
[EXTERNAL_DOWNLOADS]: The ui-analyzer subagent is designed to fetch external UI resources from design systems or guidelines via URL or MCP. This is a legitimate functional requirement for UI analysis, and the skill mandates that these analyzers inherit tool constraints from the parent environment.

subagents-orchestration-guide