agent-reach

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly asks users to "provide cookies" for logged-in platforms and delegates configuration to the agent, which implies the agent will receive and potentially embed raw cookie/auth values (secrets) verbatim — an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md 的运行路径会通过“搜索/网页阅读/视频转录”去抓取外部网页与社交内容（如 Exa 搜索结果、Jina Reader 读取任意 URL、以及社交平台帖子/评论/字幕），这些外部自由文本会被拼入 agent 的 LLM 上下文用于后续分析。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs installing rdt-cli from a Git URL with "pipx install 'git+https://github.com/public-clis/rdt-cli.git'", which fetches and executes remote code as a required runtime installation for the Reddit backend.