observability-and-instrumentation
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill demonstrates excellent security hygiene by requiring explicit user confirmation before any paid external calls, mandating the protection of secrets, and emphasizing that work should remain local whenever possible.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external repository evidence and documentation via the SandBase API, which presents a surface for indirect prompt injection.
- Ingestion points: External documentation and repository evidence retrieved via
sandbase_discoverandsandbase_inspect(SKILL.md). - Boundary markers: The workflow requires the agent to restate the requested outcome and constraints before implementation, providing a manual review checkpoint (SKILL.md).
- Capability inventory: The skill can inspect files, run tests, and apply configuration changes (SKILL.md).
- Sanitization: Verification is performed through source comparison and focused tests rather than silent execution.
- [DATA_EXFILTRATION]: Interaction with external SandBase tools is limited to vendor-owned resources (sandbaseai). The instructions explicitly direct the agent to send only the minimum necessary data and to favor existing local integrations over external calls.
Audit Metadata