The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the core agent-reach utility and several MCP servers from GitHub repositories, including github.com/Panniantong/agent-reach and github.com/xpzouying/xiaohongshu-mcp.
[REMOTE_CODE_EXECUTION]: The preflight script fetches version metadata from the GitHub API and pipes the result to a Python interpreter to extract version strings. It also uses pipx to install and upgrade the tool directly from a remote GitHub archive.
[COMMAND_EXECUTION]: The skill relies on a wide array of system commands for tool installation and operation, including pipx, pip, docker, and various CLI tools such as yt-dlp, gh, xhs, and twitter.
[CREDENTIALS_UNSAFE]: The skill manages authentication for several platforms. It provides commands for the agent to configure API keys (e.g., Groq) and authentication tokens. Notably, it includes a feature to automatically extract session cookies from local browsers using the command agent-reach configure --from-browser chrome.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion layer for data from numerous untrusted external sources, creating a potential surface for indirect prompt injection.
Ingestion points: Web content from Jina Reader and Exa, social media posts from X (Twitter), Weibo, Reddit, and Bilibili, and transcripts from YouTube and podcasts.
Boundary markers: No explicit delimiters or warnings for the agent to ignore instructions embedded in the external content are defined in the skill instructions.
Capability inventory: The skill environment allows shell execution (bash), file system modifications within ~/.agent-reach/ and /tmp/, and broad network access.
Sanitization: The skill does not document any specific sanitization or filtering logic applied to the external data before it is processed by the agent.

agent-reach