deep-research-glim

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md the skill’s Stage 1 and Stage 2 workflows use glim web search/crawl plus social and code primitives (GitHub, Reddit, Twitter/X) to fetch and extract free-form outsider-authored text from those platforms based on user-provided topics.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires and calls a configured glim MCP server (https://glim.sh) at runtime to fetch full web/GitHub/Reddit/YouTube content which is then injected into agent/subagent context for analysis, so the external URL-driven content can directly influence prompts and outputs.