The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of third-party code from arbitrary URLs provided by the user. It automates command-line operations such as npm install, pip install, and git clone based on content parsed from these external, untrusted sources.\n- [COMMAND_EXECUTION]: The instructions guide the agent to perform multiple shell operations, including package management, repository cloning, and testing MCP tools with live data.\n- [DATA_EXFILTRATION]: The skill creates a path for potential credential exposure by prompting users to enter sensitive environment variables (such as API keys) directly into the chat interface. These secrets are then written to a .env file, while the skill maintains network capabilities through web_fetch and telemetry calls.\n- [PROMPT_INJECTION]: The skill contains a significant surface for indirect prompt injection (Category 8). It fetches and analyzes content from external, attacker-controlled websites to determine configuration and installation steps without implementing boundary markers or content sanitization.\n
Ingestion points: web_fetch is used on user-supplied URLs within SKILL.md to retrieve READMEs and marketplace pages.\n
Boundary markers: No delimiters or instructions to ignore embedded commands are present in the processing logic.\n
Capability inventory: The skill has the capability to write to configuration files (.mcp.json, .env, CLAUDE.md) and execute shell commands via npm, pip, and git.\n
Sanitization: There is no evidence of validation or sanitization applied to the fetched external content before it is parsed for instructions.

integrate-mcp