planning-with-files-zh

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

• Third-party content exposure detected (high risk: 0.70). The skill's SKILL.md and templates explicitly instruct the agent to perform "查看/浏览器/搜索" operations and to write "网页/搜索结果" into findings.md, and the workflow requires immediately reading task_plan.md, progress.md and findings.md (plus a PreToolUse hook that re-reads planning files before tool calls), so untrusted web/public content can be ingested and then influence decisions/actions.