The Agent Skills Directory

[PROMPT_INJECTION]: The skill's primary function is to teach agents to adopt instructions from repository-resident files, which exposes them to indirect prompt injection. 1. Ingestion points: resources/agents-md-guide.md (Workflow 2) directs the agent to extract technical content and conventions from the project's README.md to populate AGENTS.md. 2. Boundary markers: The skill does not recommend the use of delimiters or 'ignore' instructions to isolate processed data from core agent instructions. 3. Capability inventory: The agent is instructed to execute shell commands (pnpm, npx), write configuration files, and interact with external APIs through MCP servers. 4. Sanitization: No sanitization or validation of extracted documentation is recommended before it is adopted as instruction.
[EXTERNAL_DOWNLOADS]: The skill references several Model Context Protocol (MCP) servers and IDE extensions. These downloads target well-known and trusted technology organizations, including Microsoft (@vscode/mcp-server), GitHub (@github/mcp-server), and the official Model Context Protocol project (@modelcontextprotocol/server-postgres).
[COMMAND_EXECUTION]: The skill provides procedures that require the agent to execute commands found in the project environment. This includes standard development commands like pnpm install and pnpm test, as well as npx commands for launching MCP servers. These commands serve as an execution vector for instructions defined within the codebase configuration.

ai-coding-agent-setup