The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection due to its automated investigation cycle.
Ingestion points: The skill retrieves external data such as entity fields, logs, and telemetry from a remote UModel server using umctl as seen in SKILL.md.
Boundary markers: The instructions lack clear delimiters or safety warnings to prevent the agent from following instructions that might be embedded in the retrieved monitoring data.
Capability inventory: The environment provides the ability to execute shell commands via umctl and to run dynamically generated "executable plans" for Prometheus and Elasticsearch.
Sanitization: No explicit sanitization, filtering, or escaping mechanisms are defined for the data ingested from the remote UModel server before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: Setup instructions direct the installation of the umctl tool directly from the vendor's GitHub repository at github.com/alibaba/UnifiedModel.
[COMMAND_EXECUTION]: The skill involves the execution of local setup scripts and requires the agent to perform shell command operations via the umctl CLI to interact with the monitoring infrastructure.

umodel-rca