The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install a Python package directly from a GitHub repository zip file (https://github.com/Panniantong/agent-reach/archive/main.zip). This bypasses standard package registries and executes code from a source not associated with the skill author.
[CREDENTIALS_UNSAFE]: The skill provides commands for the agent to automatically extract session cookies and authentication data from the user's local Google Chrome browser instance using agent-reach configure --from-browser chrome and yt-dlp --cookies-from-browser chrome. This represents a significant risk of credential exposure.
[EXTERNAL_DOWNLOADS]: The agent-reach install --env=auto command automates the installation of a wide array of third-party tools and CLI utilities including Node.js, mcporter, xreach CLI, gh CLI, yt-dlp, and feedparser, increasing the attack surface of the environment.
[COMMAND_EXECUTION]: The skill executes various system-level commands and binary scripts, including checking background processes (ps aux | grep xiaohongshu-mcp) and launching background services using nohup (xiaohongshu-mcp-linux-amd64).
[INDIRECT_PROMPT_INJECTION]: The skill possesses a large attack surface for indirect prompt injection as it is designed to fetch and process untrusted content from social media platforms (Twitter, Reddit, YouTube, Bilibili, Xiaohongshu, LinkedIn) and the general web.
Ingestion points: The skill reads data via xreach search, yt-dlp --dump-json, Reddit's JSON API, and Jina Reader (SKILL.md).
Boundary markers: No explicit instruction delimiters or safety warnings are provided for the agent to distinguish between control instructions and data content.
Capability inventory: The skill can execute shell commands, perform network operations via curl, and write log files to /tmp/ (SKILL.md).
Sanitization: There is no evidence of sanitization or filtering of the external data before it is processed by the agent.

agent-reach