shiny-aiconversation
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external and historical sources, which presents a surface for indirect prompt injection attacks.\n
- Ingestion points: Untrusted content enters the session via user messages, conversation search results, assistant memory entries, and text or barcodes scanned through the camera tool.\n
- Boundary markers: Historical and external data are delimited in prompts using tags such as
<conversation_excerpt>,<past_conversations>, and<memory_index>.\n - Capability inventory: The library can initiate network requests to AI providers and write data to the local file system, including SQLite databases and image attachment directories.\n
- Sanitization: The skill implements checks against path traversal in file attachments and utilizes structured data for model-tool communication.
Audit Metadata