The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for setting up various Model Context Protocol (MCP) servers and a VS Code extension (Bifrost). These resources are referenced from well-known and reputable sources, including GitHub, the VS Code Marketplace, and official package registries.
[REMOTE_CODE_EXECUTION]: Documentation includes examples of running MCP servers via npx (e.g., @vscode/mcp-server, @github/mcp-server, and @modelcontextprotocol/server-postgres). These are standard tools for the MCP ecosystem and are presented as part of a legitimate developer workflow for enhancing agent capabilities.
[COMMAND_EXECUTION]: The skill instructs the agent to perform common development tasks such as installing dependencies (pnpm install), running tests (pnpm test), and executing local project scripts for skill lifecycle management (e.g., scripts/init_skill.py). These commands are standard for codebase maintenance.
[DYNAMIC_EXECUTION]: The skill architecture includes a "meta-skill" pattern that allows the agent to capture successful workflows and generate new SKILL.md modules. This is a functional feature for project-specific automation.
[PERSISTENCE]: The guides suggest storing agent skills in global directories (e.g., ~/.claude/skills/) to ensure that specialized workflows remain available to the agent across different projects and sessions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed around the ingestion and processing of project-specific metadata (AGENTS.md). This establishes a surface for agents to receive operational instructions from the codebase context, which is the intended behavior of the AAIF open standard.

ai-coding-agent-setup