The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the Aliyun CLI tool and its associated plugins from official vendor domains such as aliyuncli.alicdn.com. These are trusted sources owned by the skill's author.
[COMMAND_EXECUTION]: It utilizes the aliyun CLI and psql for managing cloud resources (ADBPG, VPC, NAT Gateway) and database schemas. The instructions include configuring the CLI for automatic plugin installation (--auto-plugin-install true), which is a standard feature of the vendor's toolset.
[PROMPT_INJECTION]: A standard surface for indirect prompt injection exists because the skill processes learner messages and external documents (e.g., PDF or Markdown files) to provide AI-driven coaching.
Ingestion points: Untrusted data enters the agent context through learner interactions and the upload-document-async command.
Boundary markers: The skill uses structured JSON message formats (e.g., ModelParams) to separate system prompts from user input.
Capability inventory: The skill has broad capabilities including shell command execution (aliyun), cloud infrastructure management, and database operations.
Sanitization: There is no explicit content filtering or validation mentioned for the ingested documents or learner messages.
[SAFE]: The skill implements several security safeguards, including explicit instructions to never print or request full credentials (AK/SK) and providing detailed, least-privilege RAM policy templates. It also recommends using named profiles for consistent credential management.

alibabacloud-analyticdb-postgresql-ai-coaching-best-practice