microlink
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides numerous methods (metadata, markdown, html, text, extract) that ingest content from external, untrusted URLs into the agent's context, creating a surface for indirect prompt injection.
- Ingestion points: The
SKILL.mdfile defines products that fetch and parse remote web pages. - Boundary markers: There are no specific instructions or delimiters mentioned to isolate the scraped content from the agent's control flow.
- Capability inventory: The skill can perform network operations via the Microlink API and execute JavaScript in a sandboxed environment.
- Sanitization: The documentation does not describe automatic sanitization or filtering of the ingested data.
- [EXTERNAL_DOWNLOADS]: The skill suggests installing the
@microlink/mcppackage and usingnpx microlink.io. These resources are official packages from the skill author (microlinkhq) and are used for local tool execution. - [DYNAMIC_EXECUTION]: The
functionandrunmethods, along with theevaluatefield in theextractmethod, allow the agent to execute JavaScript code within Microlink's remote sandbox. This functionality is intended for advanced scraping tasks like clicking elements or waiting for specific DOM states.
Audit Metadata