kaizen-improvement-system
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including client briefs, audience/market data, campaign reports, and measurement exports (SKILL.md). This creates an attack surface where malicious instructions embedded in these documents could attempt to bypass constraints or influence the audit outcome.\n
- Ingestion points: Inputs include external briefs, content artefacts, and platform data exports as defined in the 'Required Inputs' section.\n
- Boundary markers: The system uses a 'Digital Research Currentness Gate' to verify claims and a 'Mandatory 65-to-95 Gate' which caps initial scores, creating structural limits on how ingested data affects output.\n
- Capability inventory: Capabilities are limited to 'read and search only'. Actions involving live account changes, spending, or personal-data processing require explicit, action-specific authority.\n
- Sanitization: The skill mandates the use of the 'Digital Research Engine' for source evaluation and verification to identify unsupported or fabricrated claims.\n- [EXTERNAL_DOWNLOADS]: The skill contains a reference to a 'Portfolio Kaizen standard' document hosted on the author's GitHub repository (github.com/peterbamuhigire/digital-research-skills). This reference is used to retrieve configuration and standards for the improvement cycle.
Audit Metadata