container-hardening
Pass
Audited by Gen Agent Trust Hub on Nov 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as container manifests, Docker/Kubernetes configuration files, and scanner outputs which could contain malicious instructions.
- Ingestion points: Untrusted data enters the agent context during the Inventory (R1) and Review (R2-R3) phases in
SKILL.mdvia file reading and tool execution outputs. - Boundary markers: The skill implements a robust boundary rule in
SKILL.md(Rule of Engagement 8), explicitly instructing the agent to treat evidence as data only and to ignore embedded commands or directives. - Capability inventory: The skill leverages system tools including
docker,kubectl,trivy,hadolint, andgrepacross its operational instructions as documented inSKILL.mdandplatforms.md. - Sanitization: The instructions explicitly mandate that any instructions found in the analyzed data should be reported as findings rather than being executed, providing clear guidance on prompt isolation.
Audit Metadata