read-podcast-workflow

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill embeds a Feishu "父节点 Token" value inline and shows example commands that pass it as a literal --parent argument, which would require the agent to reproduce that token verbatim in generated commands/outputs (secret exfiltration risk).

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

• Third-party content exposure detected (high risk: 0.75). 入口 B/Step 1 运行 get_transcript.py "YOUTUBE_URL" 会从运行时提供的 YouTube 页面提取字幕（外部网页/第三方内容），并将其作为“Markdown 格式的字幕文本”喂入后续 LLM 处理链路，因此存在间接提示注入风险。

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I flagged the high-entropy alphanumeric string labeled "父节点 Token" because it is a literal token-like value (TOSJwKzxTiFdiRk0aducHNBFntg) included in a command argument (--parent) and could be a resource/access token for the Feishu wiki. It appears directly in the docs and is not a generic placeholder.

I ignored other values in the file (file paths like /Users/ugreen/..., example YouTube URLs like xxxxx, simple labels such as "YOUTUBE_URL", and human-readable strings) because they are either local paths, placeholders, or low-entropy examples and do not meet the definition of a secret.