data-engineering-data-driven-feature
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input through the
$ARGUMENTSvariable, which is interpolated directly into complex instructions for various subagents (e.g., Data Scientist, Backend Architect, Frontend Developer). This represents a surface for indirect prompt injection where malicious input could influence the generated code or experimental designs. - Ingestion points: The
$ARGUMENTSvariable is utilized across multiple phases (Phases 1-16) inSKILL.mdto define tasks for subagents. - Boundary markers: Absent; user input is embedded directly within the natural language prompts.
- Capability inventory: The skill enables agents to generate architectural designs, implementation code (backend and frontend), data pipeline configurations, and analytics schemas.
- Sanitization: Technical sanitization of the input variable is absent, although the skill includes high-level procedural instructions for agents to respect data privacy and avoid recording sensitive content.
- [DATA_EXFILTRATION]: The workflow describes the integration of several well-known third-party analytics and feature-flagging services for measuring feature performance.
- External services: Integration with platforms such as Amplitude, Segment, Mixpanel, LaunchDarkly, Split.io, and Datadog to track user behavior and system metrics.
- Information flow: The skill is designed to facilitate the flow of event-based telemetry to external providers, though it explicitly instructs the agent to avoid recording sensitive session content or unconsented events.
Audit Metadata