llm-wiki-maintainer

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow ingests data by reading a user-provided “source” (files/notes/workspace slice) from the local vault and/or via MarkItDown conversion, so outsider-authored free text can be included in the material that the LLM then reads and summarizes into wiki pages.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs at runtime to create/populate the target directory from and to read control files (AGENTS.md and .skills/*/SKILL.md) in the framework repository https://github.com/Ar9av/obsidian-wiki, so fetching that external repo can directly control agent prompts and behavior.