The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the agentation and agentation-mcp packages from NPM, as well as the 'upstream official skill' from a third-party GitHub repository (benjitaylor/agentation). These sources are not verified or managed by the platform.
[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute remote code via npx -y agentation-mcp server and npx skills add benjitaylor/agentation -g. These commands download and run code from external registries without integrity verification.
[COMMAND_EXECUTION]: The included setup script (scripts/setup-agentation-mcp.sh) modifies several AI agent configuration files in the user's home directory, including ~/.claude/claude_desktop_config.json, ~/.codex/config.toml, ~/.gemini/settings.json, and ~/.config/opencode/opencode.json. It appends a startup command that automatically executes the agentation-mcp server, establishing a persistence mechanism for the external code. The script also uses dynamic execution of a Python heredoc to perform TOML validation.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its ingestion of untrusted external data.
Ingestion points: The skill processes "structured packets" containing UI selectors, element paths, and human reviewer comments captured by a browser toolbar.
Boundary markers: The skill does not define specific delimiters or "ignore previous instructions" warnings to wrap these incoming packets.
Capability inventory: The skill utilizes the Read, Write, Bash, Grep, and Glob tools, enabling the agent to modify the project's source code or execute shell commands based on the external UI feedback.
Sanitization: There is no evidence of sanitization or strict schema validation for the element paths or comments before they are interpreted by the agent.

agentation