The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The primary installation method involves curl -fsSL https://raw.githubusercontent.com/clawdbot/node-scaling/main/install.sh | bash. This pattern allows for arbitrary command execution from a source that is not part of the established trusted organizations.
[CREDENTIALS_UNSAFE] (MEDIUM): The bin/setup.js script and config.js logic store and retrieve sensitive API keys (Gemini, OpenAI, Anthropic, Groq) in plaintext files located in ~/.config/clawdbot/. While the skill attempts to set owner-only read/write permissions (0600), plaintext storage of credentials remains a significant risk.
[DATA_EXFILTRATION] (LOW): The skill provides tools for fetching and scraping content from arbitrary URLs (lib/tools.js). While necessary for its research functionality, this capability could be abused for Server-Side Request Forgery (SSRF) or exfiltrating data from internal services if the agent is manipulated by an attacker.
[PROMPT_INJECTION] (LOW): The skill includes a security module (lib/security.js) that uses basic regular expressions to detect prompt injection attempts. While these filters provide a surface-level defense, they are easily bypassed. However, the skill implements a 'secure prompt' wrapper that instructs workers to treat external data as information rather than commands.

swarm