baidu-drive

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill's commands and examples embed share passwords/extraction codes (e.g., ?pwd=abcd, -p abcd) and instruct constructing bdpan commands with those codes, which requires the LLM to insert secret values verbatim into generated commands/outputs.

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly downloads and ingests files from public Baidu share links (e.g. "https://pan.baidu.com/s/…") and its memory restore workflow (scripts/memory-backup.sh and SKILL.md "恢复记忆") fetches manifest.json and memory/workspace files from /apps/bdpan/ into the agent's workspace, allowing untrusted user-generated content to be read and materially change agent memory and subsequent behavior.

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

• Potentially malicious external URL detected (high risk: 1.00). scripts/install.sh downloads and then executes an installer at runtime from the CDN URL (https://issuecdn.baidupcs.com/issue/netdisk/ai-bdpan/installer/${VERSION}) using curl/wget and then running the downloaded installer, so remote content is fetched during runtime and executed as part of the required install flow.