podium-conversation-history-export

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests external data (conversations, reviews, and contacts) from the Podium API to build a corpus for RAG pipelines. This creates a surface for indirect prompt injection where malicious instructions embedded in conversation history could influence the behavior of a downstream AI agent.
  • Ingestion points: Data is pulled from Podium API endpoints in scripts/export_conversations.py, scripts/export_reviews.py, and scripts/attachment_downloader.py.
  • Boundary markers: The chunking process in scripts/chunk_for_embedding.py adds sender type delimiters (e.g., [sender_type] body) to the text chunks.
  • Capability inventory: The skill possesses capabilities for local file writing and network access to the Podium API for authentication and data retrieval.
  • Sanitization: The skill performs PII redaction on the emitted chunks, but does not include specific filters or sanitization for prompt injection patterns.
  • [DYNAMIC_EXECUTION]: The script scripts/chunk_for_embedding.py attempts a dynamic import of the podium_pii module to access shared redaction patterns. While this facilitates code reuse across the Podium skill suite, dynamic loading of external modules is a pattern to monitor.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:49 AM
Security Audit — agent-trust-hub — podium-conversation-history-export