review-business-kpis
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFE
Full Analysis
- [Safety Contract Enforcement]: The skill implements a robust safety contract that restricts the agent to a predefined set of read-only tools, such as
grafana_run_packandgrafana_get_evidence. This design prevents the agent from performing unauthorized write operations or accessing unintended data sources. - [Query and Input Restrictions]: The instructions explicitly forbid the agent from accepting or generating raw SQL, provider queries, URLs, or datasource identifiers. By restricting input to opaque references and scope aliases, the skill effectively mitigates risks associated with injection attacks and unauthorized network navigation.
- [Data Integrity Safeguards]: The workflow requires the agent to preserve units, severity, and timestamps exactly as returned, and it mandates prominent labeling for simulated data. These constraints help maintain the accuracy of financial and operational reporting and prevent the accidental misrepresentation of KPIs.
- [Indirect Prompt Injection Surface Analysis]: The skill ingests external data through Grafana evidence reports (SKILL.md). It addresses this attack surface by enforcing strict interpretation rules for report statuses (COMPLETE, STALE, etc.) and by limiting the agent's capabilities to a specific toolset that lacks file-writing or general-purpose network capabilities. The instructions for sanitizing errors further reduce the risk of processing malicious data payloads.
Audit Metadata