The Agent Skills Directory

[SAFE]: The skill facilitates structured UI feedback between developers and AI agents. It does not perform any sensitive file access, data exfiltration, or unauthorized command execution.
[COMMAND_EXECUTION]: The provided scripts (setup-agentation-mcp.sh and verify-loop.sh) perform standard environment setup and diagnostic checks (e.g., checking for jq, python3, and using curl for local health checks). These are transparent, well-documented, and confined to the expected functionality of an MCP server registration tool.
[EXTERNAL_DOWNLOADS]: The skill references and installs packages from trusted sources, specifically the agentation and agentation-mcp packages from the official NPM registry and a public GitHub repository belonging to the author. These are documented neutrally and follow standard development workflows.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes external data (UI selectors and human annotations). However, the architecture utilizes structured data packets and provides clear boundaries for agent consumption, which is a best-practice approach to managing indirect inputs.

agentation