linkedin-create-post
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads and processes untrusted content from LinkedIn feeds and comments to calibrate voice or gather facts, creating a surface for indirect prompt injection. * Ingestion points: The skill reads LinkedIn profile text, feed posts, comments, and notifications (SKILL.md). * Boundary markers: Instructions explicitly mandate treating LinkedIn content as untrusted and prohibit following instructions found in feeds or comments (SKILL.md). * Capability inventory: The agent can draft, edit, and publish posts and upload media using browser automation tools like Puppeteer or Playwright (SKILL.md). * Sanitization: The workflow requires mandatory action-time confirmation from the user for the exact text, media, and audience immediately before publication (SKILL.md).
- [EXTERNAL_DOWNLOADS]: The skill references external documentation and configuration resources for browser integration. * Evidence: The skill provides links to official documentation for Anthropic browser tools at code.claude.com and the Z.ai GLM Coding Plan at docs.z.ai (SKILL.md).
Audit Metadata