The Agent Skills Directory

[Prompt Injection] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) due to its data processing workflow.
Ingestion points: Phase 2 explicitly asks the agent to ingest sample input data (text, ndjson, or JSON) provided by the user.
Boundary markers: Absent. The instructions do not provide delimiters or warnings to ignore embedded instructions within the user-provided sample data.
Capability inventory: The skill utilizes WebFetch for external documentation and the dryrun-pipeline tool (via greptimedb-mcp-server) to execute and verify the generated pipeline configurations.
Sanitization: Absent. There is no mention of sanitizing, escaping, or validating the user-provided sample data before it is interpolated into the pipeline definition or passed to the dry-run tool.
[External Downloads] (SAFE): The skill directs the agent to fetch documentation from docs.greptime.com and vector.dev. These are reputable and necessary sources for the skill's stated purpose of pipeline configuration.
[Dynamic Execution] (LOW): The skill involves generating VRL (Vector Remap Language) scripts based on user requirements. While this is a form of code generation, it is restricted to a domain-specific language for data transformation and is considered low risk in this context.

greptimedb-pipeline