deep-research

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow includes “Source Discovery” and “Evidence Extraction” steps that can fetch and read public web pages or other outsiders’ documents (e.g., via web search/crawling), and then ingest their body text into the LLM context for summarization/citation.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires runtime fetching of GitHub-hosted files via the GitHub API/gh commands (e.g., gh api "repos/OWNER/REPO/contents/PATH?ref=TAG_OR_COMMIT_SHA") and MCP GitHub tools (mcp__github__*), and those remote file contents are read and injected as evidence/context that directly control the agent's outputs.