yunxiao

Warn

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill facilitates the installation of the Alibaba Cloud CLI using a remote script execution pattern from a well-known service domain.
  • Evidence: The installation process in references/install.md uses /bin/bash -c "$(curl -fsSL https://aliyuncli.alicdn.com/install.sh)" to fetch and run the setup script from the vendor's site.
  • [PERSISTENCE_MECHANISMS]: The skill includes instructions to modify system shell profiles to ensure environment variables are persisted across sessions.
  • Evidence: Instructions in references/install.md and references/reference.md guide the user to append export commands to configuration files such as ~/.bashrc and ~/.zshrc, or to use [Environment]::SetEnvironmentVariable to save tokens at the user level on Windows.
  • [DATA_EXFILTRATION]: The skill manages sensitive Personal Access Tokens (PAT) by storing them in environment variables and plaintext configuration files.
  • Evidence: The configuration flow in references/install.md requires users to provide their PAT, which is then written to the session and user-level environment variables for use by the aliyun CLI. While the skill instructs the agent not to echo these tokens, they remain accessible in shell configuration files.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and processes data from the Alibaba Cloud CLI output.
  • Ingestion points: Command outputs from tools like codeup-list-repositories and projex-search-projects are read into the agent's context as shown in SKILL.md and examples.md.
  • Boundary markers: The skill uses disable-model-invocation: true in its frontmatter, which acts as a platform-level constraint on the agent's actions during skill use.
  • Capability inventory: The skill has capabilities for command execution, file system modification (via profile updates), and network access through the CLI tool.
  • Sanitization: The skill does not explicitly define sanitization or structural validation logic for the data returned by the API before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 22, 2026, 08:52 AM
Security Audit — agent-trust-hub — yunxiao