collection
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and display information from external database environments, such as collection schemas and metrics, which could theoretically be manipulated to include adversarial instructions. The skill possesses the capability to perform destructive operations like dropping collections, though it mitigates this by instructing the agent to always seek user confirmation.
- Ingestion points: CLI command outputs for listing, describing, and querying metrics for collections as documented in SKILL.md.
- Boundary markers: Absent; there are no specific instructions to wrap external data in delimiters or ignore embedded prompts.
- Capability inventory: Command execution capabilities via the 'zilliz' CLI for creating, dropping, and modifying collections and aliases as specified in SKILL.md.
- Sanitization: Absent; the skill relies on the agent's default safety filters and the explicit requirement for human confirmation before execution of destructive commands.
Audit Metadata