classify-leads
Pass
Audited by Gen Agent Trust Hub on Mar 27, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
classify_leads_llm.pyscript is susceptible to indirect prompt injection when processing business lead data. - Ingestion points: Untrusted data is ingested from input files (typically
.tmp/leads.json) which contain information extracted from external Google Sheets viaread_sheet.py. - Boundary markers: The prompt templates in
classify_leads_llm.pylack effective delimiters or boundary markers to separate the untrusted data (name, industry, keywords, description) from the system instructions. - Capability inventory: The skill possesses file read/write access and network access to both Google and Anthropic APIs, creating a potential path for data exfiltration if the LLM is manipulated.
- Sanitization: Although the script truncates description and keyword fields to specific lengths, it does not perform any validation or escaping to prevent the inclusion of malicious prompt instructions.
Audit Metadata