ai-ad-pipeline
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as product ideas, reference videos, and user-provided briefs, to generate and distribute advertising content.
- Ingestion points: The workflow begins by ingesting a reference (Stage 0) and a
PRODUCT-BRIEF.md(Stage 1) which defines the campaign parameters and product claims. - Boundary markers: The pipeline uses structured templates for briefs and requires specific sub-skills to process different data stages, reducing the likelihood of cross-stage instruction leakage.
- Capability inventory: The orchestrator can invoke multiple sub-skills (
ad-strategy,vid-ad,static-ad,ad-distribution, etc.), write campaign artifacts to the local file system undercampaigns/<slug>/, and potentially interact with external distribution APIs. - Sanitization: The skill mandates three non-negotiable human gates: Gate A (storyboard and criteria approval), Gate B (brand, legal, and compliance sign-off), and Gate C (publish and spend approval). These checkpoints ensure that all generated content is reviewed by a human before any financial spend or public distribution occurs.
Audit Metadata