agent-dx
Pass
Audited by Gen Agent Trust Hub on Jul 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is primarily instructional and template-driven, providing a structured workflow for auditing and designing agent-facing surfaces without executing untrusted logic.
- [DATA_EXPOSURE]: The skill includes extensive guidance on preventing data exposure. Specifically, the 'sdk-telemetry' playbook mandates a content-capture toggle that defaults to off and requires boundary redaction to prevent PII leakage into observability spans.
- [PROMPT_INJECTION]: The skill demonstrates high security awareness by including heuristics to detect 'tool-description injection' (where untrusted tool metadata influences the model) and 'indirect prompt injection' risks in tool schemas.
- [CREDENTIALS_UNSAFE]: The skill provides best practices for 'credential walls' and 'delegated auth' (RFC 8693) to ensure that agent systems do not hold durable user secrets, reducing the risk of credential theft.
- [COMMAND_EXECUTION]: The skill defines a standard workflow for writing audit reports and tracking ledgers to the local filesystem (e.g.,
docs/audits/), which is consistent with its stated purpose as an auditing tool.
Audit Metadata