The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is instructed to dynamically scaffold and execute Python or Node.js scripts using Playwright or Puppeteer to crawl web pages.
[DATA_EXFILTRATION]: The instructions explicitly direct the agent to capture and log sensitive information, including HTTP session tokens, API request/response payloads, and local storage state from the target application.
[EXTERNAL_DOWNLOADS]: To perform its primary function, the skill likely requires the installation of browser automation libraries and drivers (e.g., via pip or npm) during the environment setup phase.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from external websites (DOM content, console logs, and network traffic).
Ingestion points: Web page DOM structures, interactive element metadata, console logs, and network response bodies (SKILL.md).
Boundary markers: None identified; the skill does not use delimiters or instructions to ignore embedded commands in the crawled content.
Capability inventory: The skill has access to the Bash tool to execute generated scripts and the Write tool to save reports (SKILL.md).
Sanitization: No evidence of sanitization or validation of the crawled content before it is processed or included in reports.

vibe-browser-audit