The Agent Skills Directory

[SAFE]: The content of the skill is purely documentation. It provides architectural diagrams, decision trees, and best practices for structuring AI automations. No security threats were identified across any of the analyzed categories.
[COMMAND_EXECUTION]: The skill contains a code block example of a task() tool call. This is a static instructional example for the AI to understand how to delegate work to sub-agents and does not execute arbitrary or malicious commands on the host system.
[DATA_EXFILTRATION]: The skill references standard configuration paths such as ~/.copilot/copilot-instructions.md and project-level files like AGENTS.md. These are mentioned as part of the tool's documentation on where context is stored, with no instructions to exfiltrate data from these files.
[EXTERNAL_DOWNLOADS]: The documentation references an external set of skills from a repository. This is a neutral reference for context and does not include automated downloads or execution of untrusted external code.

agent-decision-guide