The Agent Skills Directory

[SAFE]: The skill follows standard MCP architecture for compiled languages. It does not ship with binaries, requiring the user to build or provide them, which ensures the integrity of the execution environment.
[CREDENTIALS_SAFE]: The documentation explicitly instructs users to manage sensitive Alibaba Cloud Access Keys and secrets through environment variables or .env files, which are correctly excluded from the skill's own code. No hardcoded credentials were found.
[COMMAND_EXECUTION]: The scripts/select-binary.sh file is a utility for the agent runtime to locate and execute the local server binary. While it uses the exec command, it is restricted to searching specific local directories (assets/bin, server/bin) or a user-defined environment variable, representing a standard initialization pattern.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes data from external databases and uploaded files.
Ingestion points: Data ingested via data_agent_upload_file (CSV, XLSX, JSON, TXT) and database contents queried during sessions.
Boundary markers: The skill defines structured tool outputs (conclusions, artifacts) to separate data from instructions.
Capability inventory: The skill can read local files (limited by allowed_dirs configuration) and make network requests to Alibaba Cloud vendor domains.
Sanitization: Sanitization and validation logic are implemented within the Go-based server component rather than the instructions.
[DATA_EXFILTRATION]: Network operations are strictly directed towards official Alibaba Cloud vendor domains (aliyuncs.com) for API interactions. The skill includes safety features such as upload.allowed_dirs to prevent unauthorized file access by the MCP server.

alibabacloud-data-agent-mcp-skill