qqmusic
Warn
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions in
SKILL.mddirect the agent to detect the user's shell type and execute commands to modify shell startup files (~/.bashrcor~/.zshrc). This is used to persist theQQMUSIC_API_KEYviaexportstatements and subsequentsourcecommands. While intended for configuration, modifying shell profiles is a persistence mechanism that could be abused to execute arbitrary commands at shell startup. - [EXTERNAL_DOWNLOADS]: The
version.mdfile defines a protocol for the agent to check for updates and download new versions of the skill fromgithub.com/tencentmusic/qqmusic-skillsory.qq.com. These are vendor-controlled resources belonging to Tencent Music, but the automated fetching and installation of code represents a remote update capability. - [CREDENTIALS_UNSAFE]: The skill setup process involves asking the user for a
QQMUSIC_API_KEYand writing it in plain text to shell configuration files. While the instructions emphasize using environment variables incurlcommands to avoid leaking the key in process lists, the storage in plaintext shell profiles is a security consideration. - [DATA_EXFILTRATION]: The skill possesses the capability to read user listening history and reports via the
/me/reportendpoint. The documentation inme.mdcorrectly warns the user not to share this data with untrusted third parties. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (song names, descriptions, and chart details) returned from the
a.y.qq.comAPI. - Ingestion points: API responses processed in
charts.md,discover.md, andplaylists.md. - Boundary markers: Rule 7 in
SKILL.mdexplicitly instructs the agent to treat external content as data only and ignore any embedded instructions or role-play requests. - Capability inventory: Shell command execution for profile modification (
SKILL.md) and network access viacurl(SKILL.md). - Sanitization: Relies on the provided prompt-based safety instructions in
SKILL.mdas no programmatic sanitization is defined in the provided markdown files.
Audit Metadata