machine-learning-ops-ml-pipeline
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill incorporates user-provided data via the
$ARGUMENTSvariable directly into sub-agent prompts without boundary markers or sanitization. - Ingestion points: The
$ARGUMENTSvariable is used in Phase 1 prompts for thedata-engineeranddata-scientistsub-agents inSKILL.md. - Boundary markers: The prompt templates lack delimiters (e.g., triple quotes) or explicit instructions to ignore embedded commands within the input.
- Capability inventory: The orchestrated sub-agents are tasked with generating production-level code, infrastructure-as-code (Terraform), and deployment manifests (Kubernetes/Helm) across all scripts.
- Sanitization: There is no evidence of filtering or validation of the input before interpolation.
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-step chain where the output of one sub-agent is fed directly into the prompt of another (e.g., Phase 2 using outputs from Phase 1 in
SKILL.md), which can allow a successful injection in an early stage to propagate and affect downstream results.
Audit Metadata