Puppeteer

Automate Chrome and Chromium with Puppeteer for scraping, testing, screenshots, and browser workflows.

This skill appears to be a legitimate Puppeteer guide, but it includes instructions that let an agent create persistent files and store details about target sites without explicitly asking the user. Before installing or enabling it: (1) require explicit user consent before the agent creates ~/puppeteer or runs npm install; (2) inspect any scripts the agent writes to ~/puppeteer/scripts/ before executing them; (3) refuse to store credentials in memory.md — only provide secrets directly to ephemeral scripts and delete them after use; (4) consider running automation in a sandboxed account or container and restrict network access if you do not trust the skill owner (source is unknown); (5) if you want tighter control, ask the agent to show exact commands it will run (including npm install package names and versions) and to log all created files for user review. These steps reduce the risk of hidden data collection or inadvertent installation of third-party packages.

Type: OpenClaw Skill Name: puppeteer Version: 1.0.0 The skill is classified as suspicious due to its instructions for the AI agent to execute shell commands, specifically `npm install puppeteer`, `node --version`, and `npm list puppeteer` in `setup.md`. While these commands are plausibly necessary for a browser automation skill, the capability to execute arbitrary shell commands represents a significant vulnerability (potential RCE) if the agent were to be prompted to install a malicious package or execute other commands. Additionally, `setup.md` contains mild prompt injection attempts like instructing the agent to 'Don't ask — just start naturally' and to store information 'without mentioning file paths to them', which aim to influence the agent's behavior and communication style, even if not directly malicious in this context.