linkedin-personal-branding
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes an indirect prompt injection vulnerability surface through the automated ingestion of third-party web content combined with active browser interaction capabilities.
- Ingestion points: In
SKILL.md, the agent reads external, untrusted data including user profiles, activity streams, and comments viamcp__chrome-devtools__take_snapshotandmcp__playwright__browser_snapshottools. - Boundary markers: Absent. There are no instructions or delimiters specifying that the model should ignore instructions or prompts embedded within LinkedIn posts or profile fields.
- Capability inventory: In
SKILL.md, the agent controls active browser actions such as navigating pages, clicking links, and hovering over elements usingmcp__chrome-devtools__navigate_pageandmcp__chrome-devtools__click. - Sanitization: Absent. The extracted data from the web pages is processed directly without structural sanitization, escaping, or strict validation protocols.
Audit Metadata