The Agent Skills Directory

[Data Ingestion Surface]: The skill processes external telemetry and freshness evidence via tools like grafana_run_pack and grafana_get_evidence.
Ingestion points: Untrusted data enters the agent context through telemetry reports and evidence references returned by the Grafana API.
Boundary markers: The instructions do not specify explicit delimiters for the telemetry content, though they emphasize the use of opaque references and report-owned links.
Capability inventory: The skill is restricted to an observational toolset (grafana_profile_status, grafana_list_packs, grafana_run_pack, grafana_get_evidence, grafana_resolve_link) and explicitly forbids mutation, command execution, or raw query construction.
Sanitization: The workflow includes a requirement to report only sanitized errors, reducing the risk of leaking sensitive internal configuration through error messages.
[Safety Contract Enforcement]: The skill incorporates a comprehensive safety contract that proactively limits the agent's behavior. It specifically prohibits the construction of provider queries, SQL, URLs, and headers, and forbids any actions that could mutate the state of the IoT devices or the Grafana configuration. This significantly mitigates risks associated with unintended tool usage.

monitor-iot-edge