The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes MCP servers as subprocesses to facilitate tool calling. In src/client.ts, the createTransport method implements the stdio transport type by spawning a process using the command, args, and env parameters provided in the mcp-servers.json configuration file.
[DYNAMIC_EXECUTION]: The skill uses dynamic execution to manage its background processes. The cli/utils/daemon.ts and cli/commands/daemon.ts files utilize child_process.spawn to launch the MCP daemon as a detached process, ensuring it remains active across agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by design, as it enables the agent to process data from external MCP servers (such as browser-based or file-system-based servers) and act upon that data.
Ingestion points: Data from external MCP tools is returned to the agent context via src/client.ts and the call command in cli/commands/call.ts.
Boundary markers: The skill instructions in SKILL.md do not provide specific delimiters or warnings to help the agent distinguish between its core instructions and data retrieved from tools.
Capability inventory: The skill has the ability to spawn subprocesses (spawn in src/client.ts and cli/utils/daemon.ts) and perform network operations via axios and standard MCP transports in src/client.ts.
Sanitization: There is no specific sanitization logic applied to the tool output before it is returned to the agent, relying on the agent's internal safety filters.

agentic-mcp