wiki-enrich

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

• Third-party content exposure detected (high risk: 0.95). At runtime, the skill uses WebFetch to fetch outsider-authored public web content from https://alphaxiv.org/overview/<arxiv_id>.md and https://alphaxiv.org/abs/<arxiv_id>.md (and may also fetch arXiv API abstracts), then injects that fetched free text into the agent’s LLM context to generate the TODO section prose.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill fetches external source text at runtime which is fed verbatim as $SOURCE_TEXT to the LLM to generate the wiki sections — e.g. https://alphaxiv.org/overview/<arxiv_id>.md, https://alphaxiv.org/abs/<arxiv_id>.md and http://export.arxiv.org/api/query?id_list=<arxiv_id> (and it may invoke the deepxiv fetcher via python3 "$DEEPXIV_FETCHER" paper-brief <arxiv_id>), so these runtime fetches directly control prompts.