The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses StdioClientTransport in scripts/mcp-client.ts to launch MCP servers using commands and arguments defined in the .claude/.mcp.json configuration file. This allows for execution of arbitrary local binaries if the configuration is modified.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to integrate with external tools (e.g., Brave Search, Puppeteer) that fetch data from the internet. This content is returned to the agent, creating a surface for indirect prompt injection attacks.
Ingestion points: Data returned from MCP tools such as search results or web page content.
Boundary markers: The provided scripts do not implement explicit delimiters or instructions to prevent the agent from obeying instructions embedded in tool outputs.
Capability inventory: Shell command execution (via server spawning), browser automation, and file system access (via official MCP servers).
Sanitization: No custom sanitization or filtering of tool output is performed within the skill's scripts.
[DYNAMIC_EXECUTION]: The assets/tools.json catalog includes a chrome-devtools server with an evaluate_script tool, which enables the execution of arbitrary JavaScript within a browser context.

mcp-management