The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the bdpan CLI installer from official Baidu CDN domains (issuecdn.baidupcs.com) during the installation process in scripts/install.sh. It also fetches skill updates as zip packages from pan.baidu.com in scripts/update.sh. Both processes implement SHA256 integrity verification to ensure the downloaded files have not been tampered with.
[COMMAND_EXECUTION]: The skill relies on executing shell scripts and the bdpan binary to perform its core functions. It explicitly restricts the agent from using flags like --yes for updates or login, ensuring that critical operations require manual user approval.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided themes and tool outputs from the Baidu API. To mitigate potential injection risks from tool output, SKILL.md mandates that specific output blocks (demarcated by [DISPLAY_VERBATIM_START/END]) must be displayed to the user exactly as received without modification.
[CREDENTIALS_SAFE]: Authentication is handled via an OAuth 2.0 Out-of-Band (OOB) flow. The skill instructions explicitly prohibit the agent from reading or revealing the contents of the configuration file (~/.config/bdpan/config.json) which contains sensitive access tokens. Furthermore, the authorization code is passed via standard input in scripts/login.sh to prevent exposure in process lists.

baidu-wenku-aippt-personal