The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions and scripts encourage downloading and piping a remote script directly into bash (curl -fsSL https://skillhub-1388575217.cos.ap-guangzhou.myqcloud.com/install/install.sh | bash). This executes unverified code from a remote source with the current user's privileges without security oversight.
[EXTERNAL_DOWNLOADS]: References and fetches resources from an external cloud storage bucket (skillhub-1388575217.cos.ap-guangzhou.myqcloud.com), which is a non-standard distribution point for official tools.
[COMMAND_EXECUTION]: The skill relies on executing various shell commands for setup, search, and installation of external tools (skillhub install, skillhub search, Get-Command).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection vulnerabilities.
Ingestion points: The agent reads and processes output from the skillhub search command to guide user interaction (found in SKILL.md).
Boundary markers: No boundary markers or specific 'ignore' instructions are provided to the agent when processing this untrusted external data.
Capability inventory: The agent can perform write operations to the workspace (skillhub install) and execute system commands during setup.
Sanitization: There is no evidence of data sanitization, validation, or escaping for the external command output before it is parsed by the agent.

skillhub-guide