web-search

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md describes a runtime workflow where web content is fetched/extracted from outsider-provided URLs via belt app run tavily/extract and then inserted into an LLM prompt as <content> / <search-results>, creating an indirect prompt-injection path from outsider-authored page text into the agent’s LLM context.

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 invokes the inference.sh "belt" platform at runtime (e.g., https://inference.sh and its docs/ app endpoints), which runs remote apps (executes remote code and produces prompts/results the agent relies on), so this is a runtime external dependency that controls execution.