zg-ai-nine-grid-moments
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as user-provided documents, case studies, and knowledge base content to generate marketing copy.
- Ingestion points: Instructions in 'SKILL.md' (Workflow Step 1) and 'references/writing-and-evidence.md' (Section I) detail the ingestion of user-provided stories, product data, and uploaded files.
- Boundary markers: The instructions guide the agent to use square brackets for missing information and to clearly distinguish between verified facts and inferences.
- Capability inventory: The agent performs text generation and executes a local Python utility script.
- Sanitization: The skill mandates strict fact-checking against provided materials and explicitly forbids the fabrication of customer identities, dialogue, or results.
- [COMMAND_EXECUTION]: The skill utilizes a local Python script for output validation.
- Evidence: 'SKILL.md' (Step 8) specifies the execution of 'scripts/check_copy.py --max 200' to audit the final draft.
- Details: The provided script 'scripts/check_copy.py' uses standard libraries ('argparse', 're', 'sys') to perform deterministic string length and label checks without external network or system calls.
Audit Metadata