design-md

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

• Third-party content exposure detected (high risk: 0.85). Stage 4b + Stage 5: if docs_site_url is provided, the skill runs pnpm crawl:docs to fetch and render the outsider-authored documentation site into .../.claude/cache/design-md/{slug}/crawl-corpus.md, which research-collector then reads and incorporates into the LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs runtime fetches of user-supplied external sites — e.g. it executes pnpm crawl:docs "${docs_site_url}" and passes source_urls (examples include https://toss.im) into the research-collector subagent so the fetched corpus/URLs are read and injected as primary sources that directly control subagent prompts and outputs.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)