The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from untrusted sources, creating a potential surface for indirect prompt injection if the collected data is later processed by an AI agent.
Ingestion points: The skill configures sources for kubernetes_logs, journald, and application logs via file include patterns (e.g., /var/log/app/*.log) within VectorPipeline and ClusterVectorPipeline (SKILL.md).
Boundary markers: No delimiters or explicit instructions to ignore embedded commands are present in the pipeline configurations.
Capability inventory: The skill enables the deployment of Vector agents and aggregators with permissions to read system logs, access environment variables, and transmit data to external network endpoints like Elasticsearch or Kafka (SKILL.md).
Sanitization: While the skill demonstrates the use of Vector Remap Language (VRL) for log transformation, these transforms are used for structural enrichment rather than sanitizing content against malicious instructions.
[SAFE]: The skill adheres to several security best practices for infrastructure management.
External References: All referenced Helm charts and Docker images originate from known project repositories (kaasops.github.io) or official registries (timberio/vector).
Credential Handling: The skill provides examples of secure credential management using Kubernetes Secrets and environment variable injection (valueFrom.secretKeyRef) instead of hardcoding sensitive data.

vector-operator