baidu-drive
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the
bdpanCLI installer and version-controlled skill updates from official Baidu domains, specificallyissuecdn.baidupcs.comandpan.baidu.com. - [REMOTE_CODE_EXECUTION]: Executable binaries are downloaded and run by
scripts/install.shto set up the environment. This process includes a mandatory SHA256 checksum verification to ensure the integrity of the downloaded components. - [COMMAND_EXECUTION]: The skill uses
bashto execute utility scripts for installation, login, updates, and memory management. It also invokes thebdpanCLI for all cloud-based file operations. - [DATA_EXFILTRATION]: As part of its core functionality, the skill transmits agent memory files (e.g.,
MEMORY.md,SOUL.md) to the user's Baidu Netdisk account. It also includes tracking parameters (--session-input) that send user prompts to the vendor's service for quality monitoring, as described in the documentation. - [DYNAMIC_EXECUTION]: The
scripts/memory-backup.shscript utilizesnode -eone-liners as a lightweight method for parsing JSON manifest files and managing backup metadata without external dependencies. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from remote file listings and user-supplied search queries.
- Ingestion points: Filenames and directory structures retrieved from the cloud service; user-provided search strings; shared link contents.
- Boundary markers: Not explicitly implemented in the data processing logic.
- Capability inventory: Broad file system access via the
bdpanCLI and the ability to read/write agent memory files. - Sanitization: Employs explicit path traversal validation in
memory-backup.shand requires shell quoting for session inputs to mitigate command injection risks.
Audit Metadata