The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute external commands and binaries as specified in the local .claude/.mcp.json configuration file.
Evidence: scripts/mcp-client.ts uses the StdioClientTransport from the MCP SDK to spawn subprocesses using the command and arguments provided in the user-managed configuration.
[EXTERNAL_DOWNLOADS]: The skill documentation recommends the installation of several external tools and packages to provide full functionality.
Evidence: README.md and SKILL.md suggest installing gemini-cli and various MCP server packages such as @modelcontextprotocol/server-memory and @modelcontextprotocol/server-filesystem using npm or npx.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external MCP servers and presents it to the agent, creating a surface for instructions embedded in tool outputs to influence behavior.
Ingestion points: External data enters via getAllTools, getAllPrompts, getAllResources, and callTool methods in scripts/mcp-client.ts.
Boundary markers: Absent. The skill does not implement delimiters or provide instructions to the agent to ignore embedded commands in the tool outputs.
Capability inventory: The skill can spawn subprocesses (scripts/mcp-client.ts) and write to the filesystem (scripts/cli.ts writes to assets/tools.json).
Sanitization: Absent. Content from external tools is not filtered or validated before being passed to the agent.
[DYNAMIC_EXECUTION]: The skill supports tools that enable the execution of dynamic scripts in isolated environments.
Evidence: The provided assets/tools.json includes a definition for the evaluate_script tool from the chrome-devtools server, which allows the execution of arbitrary JavaScript code within a browser context.

mcp-management