The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the user with templates for SQL queries and shell commands (using curl) to analyze cluster metadata and perform data operations like BACKUP and RESTORE. These are intended for manual execution by the user and are central to the migration planning process.
[CREDENTIALS_UNSAFE]: To facilitate automated planning, the skill requests connection details including HOST_NAME, USER_NAME, and PASSWORD. It mitigates risk by providing explicit security instructions in references/sql.md, recommending the use of environment variables and .netrc files to avoid plaintext exposure.
[EXTERNAL_DOWNLOADS]: The skill references technical documentation and compatibility guides from official and well-known sources, specifically help.aliyun.com and clickhouse.com. These references are used neutrally for user guidance.
[DATA_EXFILTRATION]: The skill guides the user to collect metadata from ClickHouse system tables (e.g., system.parts, system.tables). This data is necessary for migration feasibility analysis and is not directed to unauthorized external destinations.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes cluster metadata provided by the user.
Ingestion points: Database and table metadata (names, engines, sizes) entered by the user in SKILL.md Step 3.
Boundary markers: Data is organized into structured markdown tables within provided templates.
Capability inventory: The skill has no autonomous execution capabilities (no tools or subprocess calls); it functions as a text-based advisory tool.
Sanitization: No explicit sanitization of database metadata is mentioned, but the lack of dangerous capabilities mitigates the risk.

alibabacloud-cksync-plan