The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple zilliz CLI commands (e.g., zilliz context current, zilliz cluster describe, zilliz database list) to retrieve environment information. These commands are consistent with the skill's stated purpose of providing a status overview and use the official vendor utility.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data returned by the Zilliz CLI, such as database names and collection statistics, which could theoretically contain malicious instructions if the environment were compromised.
Ingestion points: Output from Step 2 through Step 5 in SKILL.md is fed back into the agent's context for summarization.
Boundary markers: The instructions explicitly request --output json, ensuring the agent processes structured data rather than unstructured text, which helps maintain context boundaries.
Capability inventory: The skill is limited to read-only metadata retrieval and status checking using the zilliz CLI tool.
Sanitization: No explicit sanitization is performed, but the risk is minimal as the data originates from the user's own authenticated cloud environment.
[SAFE]: All operations performed are standard for a cloud environment management tool and are restricted to the vendor's official command-line interface.

status