melech-market-validation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as interview transcripts, support tickets, and public forum posts.
- Ingestion points: Data is ingested through 'mine first-party evidence' (Step 3) and 'primary discovery' (Step 6) workflows in SKILL.md, as well as specialized 'research lanes' in references/orchestration.md.
- Boundary markers: Instructions in references/artifacts.md recommend keeping facts, statements, and interpretation separate in reports. references/orchestration.md requires evidence lanes to return data in a specific JSON schema, which acts as a structural boundary.
- Capability inventory: The skill utilizes web research, file-system access to read workspace files (e.g., support logs, CRM data), and multi-agent orchestration.
- Sanitization: The skill advises using aggregated or redacted evidence (Step 3) and requires limitations to be noted for every evidence row, but lacks explicit escaping for data interpolated into agent prompts.
Audit Metadata