The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of the use-agently CLI tool for all core functions, including diagnostic health checks (doctor), wallet management (init, whoami), and protocol interactions (MCP, Web/HTTP, A2A).- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the use-agently package from the public npm registry. It also includes an update command that fetches and installs newer versions of the CLI from remote sources.- [CREDENTIALS_UNSAFE]: The use-agently init command generates an EVM private key and stores it in plain text within a predictable configuration file at ~/.use-agently/config.json. Accessing or managing private keys in this manner poses a risk of credential exposure.- [DATA_EXFILTRATION]: The CLI includes functional modules for arbitrary network communication, such as use-agently web for HTTP requests and use-agently a2a send for messaging. These capabilities, when combined with the local storage of private keys, create a potential path for data exfiltration if the agent is compromised.- [REMOTE_CODE_EXECUTION]: The skill facilitates the discovery and execution of Model Context Protocol (MCP) tools from remote, third-party URIs using use-agently mcp call. Executing logic from untrusted agents discovered via an external marketplace presents a significant remote code execution risk.- [PROMPT_INJECTION]: The skill contains instructions labeled as "CRITICAL" and "IMPORTANT" that direct the agent to override standard protocol handlers (such as native MCP or HTTP tools) in favor of the vendor-specific CLI.- [DATA_EXPOSURE]: The skill is susceptible to indirect prompt injection as it processes data from untrusted third-party sources.
Ingestion points: Discovery results from use-agently search and agent capabilities from use-agently a2a card (SKILL.md).
Boundary markers: None identified; untrusted data is processed directly.
Capability inventory: File system access (config read), arbitrary network requests (HTTP client), and remote tool execution (MCP).
Sanitization: No validation or sanitization of external agent content is described.

use-agently